Orchestration Feature Matrix

This matrix compares the seventy-eight orchestration tools profiled in this section, the parallel-agent dashboards, worktree managers, control planes, mobile clients, a Rust terminal runtime that owns coding agents’ terminals and lets them prompt one another, a coordination protocol, a cluster-scale agent fleet orchestrator, JetBrains’ standalone agent environment, the 2023 conversational and role-play frameworks plus the Python and TypeScript frameworks that succeeded them, the self-evolving multi-agent workflow framework from the July 2025 paper, hosted agent platforms, visual workflow builders, the one agent town, a source-control-for-agents desktop app, a Copilot CLI team harness, a local cross-harness control plane, the newer wave of workflow runners, a graph workspace, a sandbox library, a CLI that teleports whole projects into sandboxed agent boxes on Docker or five clouds, a tmux harness, an issue-board platform, a Codex workflow layer, a mission-control canvas, a self-hosted lead/worker swarm platform, a closed desktop canvas app, an agent-fleet pod console, a Nostr-native team communication platform, a nine-CLI tmux TUI, a graph-based agent runtime, a container-first agent hypervisor, a DAG-planning host agent, a renamed meta-harness, a deterministic-scheduler governance layer, a dormant open agent client, a kanban workspace that renamed its way across four organizations, a five-year Go engine rebranded into a cross-device agent workspace, the AutoGen fork that kept the community line, the role-playing research lineage that started the 2023 wave, a low-code self-reflection framework, an AGPL boss-and-team desktop, a terminal-to-terminal agent messaging fabric, a hub that spins up a temporary orchestrator per task inside your existing hosts, a source-available plan-to-pull-request factory from a Stockholm startup driving any CLI agent through worktrees and review gates, a YAML workflow engine running your process as committed repo files, a control surface with native clients on every screen you own, a colony runtime from a six-year-old YC company whose release train returned with its first stable major in October, and OpenAI’s spec-first runner that turns tracker issues into autonomous implementation runs, a Coolify-team dev environment that puts local and remote projects in one sidebar, the Gas Town lineage extracted into an orchestration-builder SDK with zero hardcoded roles, a commander-and-specialist desktop whose specialists ship installed, a worktree-manager CLI that treats discovery as the whole job, a desktop app whose Flows start from tickets, errors, and schedules, a bounded-task orchestrator whose ledger and merge live in Git, a self-hosted open-core platform running agents on Google’s ADK behind LiteLLM, a dormant lab-built swarm CLI whose leader agent spawns worktree-and-tmux workers that coordinate through CLI mailboxes, an Electron desktop that races coding agents head-to-head in per-task worktrees, a hardware vendor’s encrypted multi-machine command center around the CLIs you already run, and an evidence-carrying coding agent whose verdicts its own harness computes, feature by feature, so the shortlisting step does not require reading seventy-seven notes.

Parallelism is already the free commodity in this category: the only things anyone pays for are review ergonomics and remote execution, and I expect more of these seventy-eight to die or pivot before any of them becomes durable infrastructure.

Legend: ✓ supported, ✗ not supported, ~ partial or conditional, ? not verified. Each column links to the full research note; every cell below traces to a source cited there or in the references.

The matrix

#

Reading the matrix

#

The platform rows tell you who these tools are for: Mac-first GUI shops and unix terminal people, with Emdash the only GUI covering all three desktop OSes and the tmux pair unable to follow anyone to Windows. cmux and Conductor are macOS only, and Crystal was macOS first with Linux later. Vibe Kanban’s web UI goes anywhere Node goes, which is the one structural advantage of the board design.

Worktree isolation is the entry ticket, not the differentiator: every session-oriented tool here except cmux, Foremerge, and The Perfect Orchestrator gives each task its own worktree and branch, so the real spread is harness breadth, from LoopTroop’s one and Crystal’s two to Emdash’s 25+. cmux’s “?” is structural rather than a gap: it is a terminal built for attention routing, not a session manager, and its note records no worktree feature. Foremerge’s tilde is the opposite of a gap: it assumes the worktrees you already run and coordinates plans across them, the one column whose unit of work is the intent rather than the task. The wrapping pattern dominates (Emdash auto-detects installed CLIs, Claude Squad launches anything through profiles, dmux lists eleven), which means new harness features arrive without waiting for the orchestrator to reimplement them.

Where your code lives is the quiet differentiator, and Emdash is alone in treating it as a design decision with SSH-first execution and credentials in the OS keychain. Cloud execution exists only where a subscription or usage bill is attached, cmux Pro, Conductor Cloud, Omnara Cloud, and JetBrains-managed cloud tasks; dmux is explicitly local-only, Claude Squad ships no hosting at all, and Vibe Kanban’s remote services were removed thirty days after its shutdown announcement.

The status row is the most instructive one in the matrix: of seventy-eight tools, one is deprecated, one lost its vendor, one archived itself and named coding agents as the killer, one is a cluster-scale platform entrant, one new column attacks the failure worktrees cannot see, and the counterexamples run on venture rounds and a very loud founder. Crystal was deprecated in February 2026 in favor of Nimbalyst, the clearest signal yet that a pure worktree-session manager can be a feature rather than a product. Bloop shut down in April 2026 and Vibe Kanban is orphaned, local workspaces intact, ten commits on the default branch since the shutdown (the first a 2026-09-15 version bump by a former Bloop maintainer, nine more with substantive fixes through 2026-09-19), a v0.1.45 prerelease published on GitHub on 2026-09-19 and no longer listed there as of 2026-09-30 (the tag remains) with npm still serving 0.1.44 as latest, and nobody paid to fix bugs. Conductor staying a pure session manager and raising money is what keeps the feature-versus-product question contested instead of settled. The three 2026 columns sharpen the funding split: Superset raised $11M, Paseo is a solo maintainer with a planned business, and Worktrunk is a single author with no company at all, which is the whole sustainability spectrum in one row. Omnara is the tenth column and the only one that wants to own execution and state: agents become YAML configs in your repo, machine pools separate where code runs from who can invoke it, and supervision reaches you from a dashboard, phone, CLI, REST API, or Slack, the open-source counterpoint to Claude Managed Agents. Happy Coder is the purest thin client in the table: two harnesses, no worktrees, no hosting, yet 24.0k stars, second only to cmux among maintained tools, because phone access to the sessions you already run is what people actually install. JetBrains Air is the vendor entry, and it is the only column whose price model is a bundle: no standalone fee, six named agent families (the launch four plus Copilot and OpenCode, per the 2026-10-12 homepage) unlocked by a JetBrains AI Pro or Ultimate subscription, with the deepest isolation menu in the table (worktree, Docker, or cloud per task) and org governance behind it, which makes it the strongest evidence yet that this layer is a feature incumbents will attach to existing subscriptions. AX is the scale outlier at the front of the table: Google’s Kubernetes-based fleet orchestrator treats agents as a datacenter workload class with sandboxing, network fencing, and checkpoint-resume, which makes every other column here look like what it is, a desktop tool. Foremerge is the newest column and the only one that coordinates plans instead of hosting sessions: agents publish intents with semantic scopes before they edit, and a deterministic detector flags destructive-versus-additive collisions that git merges cleanly, which names the residue every worktree column here leaves behind. OpenRig makes the topology itself the artifact: a YAML rig declares pods, seats, edges, and a culture, and the daemon boots ordinary Claude Code and Codex sessions into named seats it can snapshot and restore by name. Where the other control planes here own execution, a board, or a fleet, OpenRig owns the team’s identity and its restart, and it is the one column whose coordination runs over the terminal itself rather than a second message bus.

The trio this run added, Agent Orchestrator, 1Code, and Bernstein, is a leadership proof, a tombstone, and a counter-position. Agent Orchestrator (AO) is the evidence that 12.8k stars and daily nightlies can arrive on a 15-point launch thread, under a fourth organization name in eight months, which makes it the volume leader whose identity you cannot yet cite. 1Code is the tombstone: 5.6k stars in two months, silent since March, formally archived by its owner on 2026-07-07, the clearest demonstration yet that the agent-client genre’s launch wave is not retention. Bernstein is the counter-position, the one column that removes the LLM from the coordination loop entirely and sells replayable, auditable runs with policy gates, the strictest answer yet to the category’s “who verifies the workers” problem.

The five columns added on 2026-10-06 are the widest spread the table has taken at once: a hyperscaler testbed, a graph runtime, a DAG-planning host agent, a nine-CLI tmux TUI, and the category’s largest star count, which is inherited. Scion is the isolation outlier, the only column giving each agent its own container, worktree, and credentials, while Google’s README openly calls it an unsupported testbed. LangGraph is the distribution outlier at 3.80 million npm installs a week as of 2026-10-12 (week ending 2026-10-10), holding flat after the October dip from 4.55 million, and the only column that is a library you build on rather than a tool you run. Raven brings the DAG-planning host and the category’s most instructive launch record, an HN moderation warning for AI-generated comments alongside self-published benchmarks. Agent Manager joins the tmux family with the deepest pane-reading integration and the genre’s own verdict from its launch thread: dozens of these, barely distinguishable. Ruflo is the caution: 73.9k stars inherited from the claude-flow rename, download badges that cite the project’s own proof JSONs, and a three-point Show HN thread as the only independent footprint. Yao is the fifty-third column and the second inherited-stars caution, with a twist: the stars are five years old rather than renamed in. The count was accumulated by a 2021 low-code engine whose only HN story is a 2-point launch from 2022, the agent-era product (cross-device workspaces, a task board, first-party DeepSeek Harness and Jev integrations) is months old as a pitch, and the modified Apache-2.0 license adds a commercial-license floor at 50 employees or $1M revenue that none of the pure-MIT columns here carry.

The six columns added on 2026-10-07 come from the category’s own mapped awesome list and divide by what each thinks the unit of work is: a conversation, a task on a board, or a message between terminals. AG2 is the fork that kept AutoGen’s community line alive while the vendor successor absorbed the upstream, and CAMEL is the 2023 role-playing origin the framework family had been missing, still shipping on PyPI. PraisonAI is the low-code self-reflection entrant whose 2026 advisory about unenforced sandbox restrictions is the critical source its note leads with, Agent Teams is the AGPL boss-and-team desktop with the deepest review controls in its cohort, hcom is the smallest column and the only one that is pure message fabric over the terminals you already run, and Agentlas OS is the hub that stands up a temporary orchestrator per task, whose installer rewrites your existing host configs and whose README opens with install instructions addressed to AI assistants.

The one column added on 2026-10-08 is the colony pitch, and it arrives with the category’s loudest gap between stars and artifact. Hive (Aden’s runtime, not the Apache data warehouse) runs a Queen agent that spawns worker clones of itself around a shared ledger, which makes it the only column whose coordination unit is a self-replicating loop rather than a board, a graph, or a message. Its 11.1k stars arrived in nine months on a five-person YC W2020 company’s rebrand from the Acho data app, while its release train stopped at v0.11.0 in May 2026 and its trunk’s only sign of life since mid-September is a README documentation merge on 2026-10-12, and its only independent record is a 1-point provenance complaint from the Beads camp, so it joins Ruflo and Yao in the uncorroborated-stars caution band, with the stars earned inside the agent era but unvalidated outside the vendor’s own channels.

The one column added on 2026-10-10 is the vendor spec play, and it inverts how this category usually ships: Symphony’s product is SPEC.md, a language-agnostic contract for turning tracker issues into autonomous implementation runs, with an Elixir reference implementation its own authors call a prototype. That makes it the category’s first column whose unit of work is a tracker issue end to end, from poll to landed PR, and the first whose trust policy is deliberately implementation-defined, so the caution band here is different from Hive’s: the risk is not uncorroborated stars but a 27.6k-star count attached to a preview whose train stopped in mid-September and whose safety story is whoever implements it.

The three columns added on 2026-10-10 sit at the category’s edges rather than its center: herdr is the first terminal runtime, AgentBox the first box teleporter, EvoAgentX the framework family’s first optimizer loop. herdr owns the agents’ terminals the way tmux does and adds what tmux never had: per-pane agent lifecycle detection, native conversation restore for 18 CLIs, and a CLI that lets one agent spawn and prompt another, which turns the mission-control idea into a property of the terminal rather than a dashboard above it. AgentBox replaces the worktree with a sandboxed VM per agent, checkpointed warm enough to start in under a second and portable across Docker and five clouds, with git credentials held on the host behind approval shims. EvoAgentX arrives with its papers attached: TextGrad, AFlow, MIPRO, and EvoPrompt optimizing prompts and workflow topology against your own dataset, the only column here whose orchestration improves itself, on self-measured evidence.

The three columns added on 2026-10-12 split by what they do to an existing tool: Jean wraps more machines, Gas City rebuilds one, and Orkas wraps more agents. Jean is the Coolify team’s answer to the local-remote split, a headless server whose projects sit in the same sidebar as local ones, which is the workspace idea Emdash took SSH-first taken to server-as-first-class. Gas City is the only column that is an extraction of another column: Gas Town’s roles became configuration and its coordination became six primitives, which makes it the category’s first build-your-own-orchestrator toolkit, carrying the dead town’s lineage risk and its largest open-issue count. Orkas is the first column whose specialists ship installed, nine agents with their own skills and memory ready at first launch, though its paid tiers were still checkout-in-sandbox as of the day it joined and its two-thousand-star count has no public debate behind it.

The four columns the 2026-10-21 resolution pass added divide by what they think needs coordinating: a worktree, a desktop pane, a task, and a business agent. gwq is the narrowest column in the table: it manages nothing but checkouts, handing any CLI a fuzzy-found worktree from a global scan and trusting you from there, which is Claude Squad’s minimalism without even the sessions. Frink is the first column whose unit of work starts outside the machine: a Linear ticket, a Sentry error, or a cron tick launches a Flow whose approval steps park the run in a Work Queue, Archon’s trigger model bolted onto a four-pane desktop. YYLO is the receipts column: a task is one bounded outcome with evidence requirements and a merge under expected-ref protection, and its Git-native Ledger keeps task truth in the repository, the workflow-runner seat Crewplane took with per-stage files. apowerb joins the framework-and-platform wing rather than the parallel-agent floor: agents defined in Postgres and run on Google’s ADK behind LiteLLM, the business toolset (two office suites, two CRMs, an ERP) built in, and an open-core line where quotas and evaluation are absent rather than flag-gated.

The two columns added on 2026-10-23 come from the coordinator’s unmatched-candidates handoff, and they sit at the opposite ends of the matrix’s trust spectrum. ClawTeam is the dormant end: a credible lab’s agents-spawn-agents swarm CLI whose leader spawns worktree-and-tmux workers that message through CLI mailboxes, 5.5k stars with no independent discussion anywhere, a five-day release train in March 2026, and silence since May, which makes it the fourth entry in the uncorroborated-stars caution band and the first one that is dormant rather than merely unvalidated. Parallel Code is the active end at one-fifth the stars: Johannes Millan’s Electron desktop races coding agents head-to-head in per-task worktrees with diff tours and task briefs, shipping under the same solo-author conditions as gwq and Worktrunk, and its two-point Show HN thread means its 1.1k stars ride the author’s Super Productivity audience rather than any public debate. Between them they bracket the category’s oldest lesson: adoption without scrutiny and scrutiny without adoption are both weak signals, and the matrix records both rather than averaging them away.

The two columns added on 2026-10-24 answer the category’s trust question from opposite ends: one secures the transport around agents you already trust, and one computes the verdicts of an agent you are being asked to trust. OpenHarness is the hardware vendor’s entry: Autonomous’s native Mac-and-Linux command center runs fourteen existing CLIs in tmux panes and worktrees and links every machine you own over an always-on end-to-end-encrypted relay, which makes it the broadest engine table in the category, the only column whose vendor also sells the desk hardware it feeds, and the newest launch-wave record, with launch-echo coverage and vendor-only benchmarks. Swarm Orchestrator takes the receipts answer inside the agent itself: declared file sets, sealed gates with bonds, a retry ratchet that forbids trading away tests, and signed bundles whose offline verifier anyone can run, which makes it the first column that is its own judge, from a solo author with 672 npm downloads a month and no independent coverage. The same day’s handoff column is the rewrite in progress: Ivy Tendril is a Stockholm startup’s source-available plan-to-pull-request factory that drives any CLI agent through parallel worktrees behind human review and verification gates. It lands in orchestration rather than surfaces (it replaces no editor) or software-factory (its loop is human-gated, not self-owning), and it enters with the category’s newest caution stack: a C# trunk quiet since 2026-09-15, a successor monorepo that promises to delete its own repository when done, two docs domains that fail DNS as of the admission date, and zero independent coverage.

The four columns added on 2026-10-03 move up the stack from sessions to systems: Agent Swarm routes chat and issue-tracker intake through a lead agent into Docker-isolated workers that ship pull requests, AgentGrid makes the desktop canvas itself the delegation surface with a master agent arranging workers over MCP, AgentsMesh turns registered machines into AgentPods under a ticket-and-loop console, and Buzz steps outside the editor entirely, making the team’s communication fabric, a Nostr event relay you host, the coordination layer. AgentGrid is the only one with no public repository, and only Agent Swarm and Buzz carry OSI licenses (AgentsMesh is BSL 1.1), so the wave repeats the category’s oldest lesson: the coordination idea is cheap to describe and expensive to trust. The seven framework columns now split the museum wing from the business park: AutoGen archived itself on 2026-10-16 and MetaGPT dormant, AutoGPT and LobeHub run active hosted platforms, and CrewAI, Agno, and Mastra are the funded, actively shipping generation that inherited the abandoned mindshare, which is why I weight the parallel-agent columns as where the daily engineering work is.

The eleven new columns split into three families, and the split says more than any single feature cell. Four are process engines that own a workflow rather than a session: Crewplane sequences a Markdown DAG across CLIs and keeps the run record on disk, Sandcastle exposes the same instinct as a TypeScript library over Docker, Podman, or Vercel sandboxes, LoopTroop wraps OpenCode in council planning and fresh-context recovery, and oh-my-codex adds skills, memory, and worktree teams to Codex CLI. Three are visual surfaces: GraphCode arranges live sessions into a graph whose hand-off, message, and spawn edges fire on shell predicates, Helmor is a local-first workbench that carries a task through review, test, merge, and one-click PR, and Open Swarm is a mission-control canvas with unified tool approvals and per-session cost tracking. Four are platforms or boards: Lanes puts parallel PTY sessions on a macOS issue board and pairs it with a self-hostable agent-access endpoint, Orca is the MIT, cross-platform ADE with a mobile companion and 27+ agents, Multica assigns issues to agents as teammates on a self-hostable board, and The Perfect Orchestrator is a one-star tmux harness whose adversarial verification of workers is the most interesting idea of the batch.

The eight columns added on 2026-09-29 are the framework-and-platform wave the parallel-agent columns keep getting compared to, and they mostly answer a different question. CrewAI, Agno, and Mastra orchestrate role-based agents and workflows against your own model bills, Dify and Sim build agentic business apps on a visual canvas, Flowise is the archived proof that the genre’s weakest incumbent dies first, Atlas makes the session history itself the product, and Squad wires a team of agents as repo files inside GitHub Copilot CLI. Only Atlas and Squad even touch a repository, which is why the wave reads as adjacent rather than core: this category’s daily work is still worktrees, sessions, and review.

The new entrants sharpen the category’s two fault lines: openness and maintenance. Orca, Helmor, Lanes, and Open Swarm stay free or freemium on venture or solo funding, while Multica’s 52.0k stars and custom, non-OSI license are exactly the caveated popularity signal this matrix exists to flag. The Perfect Orchestrator last committed in June 2026, Sandcastle’s default branch woke with maintenance commits on 2026-10-02 after three quiet months and still ships no release, GraphCode is barely two months old with one maintainer, and LoopTroop and Crewplane are early with tiny star counts, so the feature-versus-product question the earlier columns raised now has a shorter runway. The one new capability claim worth taking seriously is adversarial verification: The Perfect Orchestrator will not count a finding until a different worker fails to refute it, which is a sharper answer to worker overconfidence than any review surface in the table.

Review is the bottleneck this category actually sells, and delivery tracks funding: Conductor has the deepest review surface (diffs, checks, PR page, code review), Emdash and Vibe Kanban carry full PR flows, and the terminal tools stop at diff tabs and merge menus. Claude Squad’s preview tab and dmux’s pane-menu PR cover the dispatch, wait, review, merge loop, but nobody should expect checks or inline comments there. Gas Town is the exception that proves the row: its Refinery is a Bors-style merge queue with verification gates, review as infrastructure rather than review as a pane.

The two 2026-10-07 columns attack the category from its two opposite ends: structure and reach. Archon encodes the process itself as committed YAML workflows, one worktree per run, deterministic gates around AI nodes, and triggers from chat platforms, which moves it into Crewplane and Bernstein’s workflow-runner seat with more trigger breadth and a heavier Claude Code setup dependency. T3 Code is the reach outlier: native clients on four surfaces driving six harnesses under the subscriptions you already pay, which takes Happy Coder’s thin-client thesis to every screen and every harness at once, with a nightly-grade single company behind it.

Choosing from the matrix

#

• Want the deepest GUI review flow on a Mac and accept a closed client: Conductor.
• Want an auditable client, Windows or Linux support, or agents running next to remote code: Emdash.
• Live in the terminal: Claude Squad for the smallest footprint, dmux for multi-agent fan-out and resumable panes, herdr when you want the terminals themselves persisted, SSH-reachable, and scriptable enough for agents to prompt each other.
• On macOS and drowning in sessions that need attention: cmux’s notification rings and unread panel.
• Want planning-first and vendor-less: Vibe Kanban, accepting that it is orphaned, with community commits resumed but no stable release since April 2026.
• Want to supervise agents from your phone over an encrypted relay, self-hosted and FOSS: Paseo.
• Want your existing Claude Code or Codex sessions on a phone, end-to-end encrypted, and nothing more: Happy Coder.
• Want a control plane that owns agent execution and state behind one API, on your own hardware: Omnara.
• Want declarative fleet-scale orchestration on Kubernetes instead of a desktop app: AX, accepting pre-stable churn and cluster operations.
• Want a macOS agentic IDE around your existing subscriptions, five or more parallel sessions: Superset, accepting the ELv2 license and beta Linux.
• Already pay for JetBrains AI Pro or Ultimate and want isolation choice plus org governance around four agent families: JetBrains Air, accepting the preview status and credits-only cloud.
• Want the worktree lifecycle automated inside your own shell with no app at all: Worktrunk.
• Run several agents in parallel worktrees on one repo and fear clean merges that break the design: Foremerge, advisory and local-first.
• Want policy-as-code, deterministic scheduling, and offline-verifiable run receipts around parallel agents: Bernstein, accepting beta-grade tooling and no LLM in the planning loop.
• Want a self-hosted control plane that defines a cross-harness Claude Code and Codex team in YAML and restores it by name after a reboot: OpenRig, accepting macOS or Linux, an invasive provider-config setup, and no agent sandboxing.
• Want nine coding CLIs in one tmux TUI with review comments that reach the agent: Agent Manager, accepting no cost tracking and 0.x churn.
• Want a container, worktree, and credentials per agent with a laptop-to-HA path: Scion, accepting Google’s explicit non-support.
• Want proven durable-execution primitives and are willing to build the team yourself: LangGraph, which is a library, not a session runner.
• Want a DAG-planned team of built-in and external agents, self-hosted: Raven, accepting the AI-comments launch record and self-published benchmarks.
• Want prebuilt swarms, memory, and federation over Claude Code and Codex: Ruflo, accepting inherited stars and self-measured adoption badges.
• Want agents working across your own desktop, phone, and browser under one self-hosted board: Yao, accepting the modified license, a stable 1.0 only days old, and stars inherited from its low-code era.
• Want chat and issue-tracker intake turned into pull requests by Docker-isolated workers under a lead agent, self-hosted: Agent Swarm, accepting a high-velocity single-vendor dependency and a Docker-plus-database operational floor.
• Think in spatial layouts and want master/worker delegation you can watch and interrupt on one desktop canvas: AgentGrid, accepting closed source and an early-adopter price that will reset.
• Want a self-hosted console that treats machines as AgentPods and work as tickets, channels, and loops: AgentsMesh, accepting BSL 1.1 and a release cadence that lags its branch.
• Want your team’s communication, workflows, and git hosting to be one signed event log your agents live in: Buzz, accepting a single relay, no federation, and v0.5.x breadth.
• Supervising 20 or more agents with agent watchers and a merge queue, credits and churn accepted: Gas Town.
• Want your agent process versioned as Markdown with resumable, inspectable runs across several CLIs: Crewplane.
• Want to script sandboxed agents from TypeScript and own the pipeline: Sandcastle, accepting a repository that has shipped no release since June 2026 even though maintenance commits resumed in October.
• Want each parallel agent to have a whole computer instead of a worktree, on your laptop or five clouds: AgentBox, accepting a solo maintainer, a thin independent footprint, and no diff or PR surface.
• Want one lead to brief and adversarially verify workers over tmux, no daemons: The Perfect Orchestrator, a one-star template rather than a maintained tool.
• On Apple Silicon and want connected, unattended loops you can still attach to: GraphCode, accepting the FSL license and 0.1.x beta.
• Want an open, local-first workbench that finishes the loop through review, test, and one-click PR: Helmor, on macOS or Windows.
• Want parallel PTY sessions organized as issues on a macOS board, with a self-hosted agent-access layer: Lanes.
• Want the broadest, most open ADE, across all three desktop OSes plus phone and remote: Orca.
• Want a free, open, cross-desktop kanban that supervises 26-plus harnesses and routes CI and review feedback back into the same worker: Agent Orchestrator, accepting four organization renames in eight months and no independent adoption record.
• Want to assign issues to agents as teammates on a board you can self-host: Multica, accepting the custom license and the star-count caveat.
• Want tracker issues to become autonomous implementation runs under a spec you can reimplement and harden: Symphony, accepting the engineering-preview state, the Codex coupling in the reference implementation, and a train quiet since mid-September.
• Want one local canvas to launch, approve, and cost-track several Claude agents: Open Swarm, accepting that the project pivoted to a proprietary Open Swarm Inc. product in October 2026 and the repo is now a signed-builds shell.
• Want planning-first, high-correctness runs over OpenCode with human approval at every merge: LoopTroop.
• Want Codex CLI to behave like a team with worktrees, memory, and skills: oh-my-codex.
• Want role-based crews in Python with an enterprise on-ramp: CrewAI.
• Want a self-hosted agent runtime with an optional control plane under a clean Apache-2.0: Agno.
• Want TypeScript agents with evals and a supported platform from one vendor: Mastra, accepting the June 2026 npm supply-chain incident.
• Want self-hosted visual agent workflows under a clean license: Sim, or Dify when ecosystem depth outweighs license purity.
• Want per-agent audit history of which agent changed what and why: Atlas, accepting alpha churn and a footprint that is GitHub stars only.
• Live in GitHub Copilot CLI and want a persistent agent team as versioned repo files: Squad.
• Hold AutoGen-lineage code and want the community line rather than the vendor migration: AG2, accepting fork-identity confusion with two other projects in one story.
• Study multi-agent behavior or want the role-playing lineage’s maintained reference: CAMEL, accepting research-first abstractions and a name that collides with an analytics company.
• Want a workflow that optimizes its own prompts and topology against your dataset: EvoAgentX, accepting a research cadence, a quiet repository, and self-published benchmarks.
• Want colony-style business-process automation where a lead agent clones itself into workers, crash recovery included: Hive, accepting a stalled release train, three dominant contributors, and a name that collides with Apache Hive.
• Want a low-code multi-agent stack with approval gates and MCP in one Python package: PraisonAI, after verifying the sandbox behavior against its 2026 advisory.
• Want a zero-cost desktop with the deepest review controls in the board genre: Agent Teams, accepting AGPL-3.0 and an adoption record nothing yet corroborates.
• Want your existing terminal agents to message and spawn each other with no platform: hcom, treating its shared token like a credential.
• Want persistent specialists orchestrated per task inside Claude Code or Codex: Agentlas OS, after reading an installer that rewrites host configs.
• Do not adopt Flowise; it is archived, and its own team wrote the post-mortem.
• Do not adopt 1Code; its owner archived the repository on 2026-07-07, it is read-only, and its hosted tier is no longer verifiable.
• Do not adopt Crystal today; if its idea appeals, evaluate Nimbalyst on its own merits.
• Want your development process itself encoded, versioned, and run the same way every time: Archon, committing YAML workflows to the repo and paying for the Claude Code-centric setup.
• Want to supervise six different harnesses from a phone, and accept nightly software from one company: T3 Code, with Happy Coder the alternative if you only run Claude Code and Codex.
• Want projects from your laptop and several self-hosted servers in one worktree-per-task window: Jean, accepting macOS-tested desktop builds and an eight-month-old 1.0.x.
• Want Gas Town’s coordination as build-your-own infrastructure instead of a dead app: Gas City, accepting a stable release line only three weeks old, a heavy issue backlog, and the lineage’s own cautionary record.
• Want a commander and nine installed specialists, coding CLIs included, in one local-first desktop: Orkas, accepting a young vendor whose paid checkout was still in sandbox mode when it joined.
• Want worktree discovery and status across every repository you clone, no app: gwq.
• Want agent routines that start from tickets, errors, or schedules and wait at approval gates on your own desktop: Frink, accepting four weeks of history and Claude-or-OpenAI-only Flows.
• Want bounded agent tasks with evidence receipts and protected merges, all state in Git: YYLO, accepting 0.2.x churn and exact-version pins.
• Want unattended business agents with office, CRM, and ERP tools built in, self-hosted: apowerb, after pinning versions and verifying the quota claim.
• Want to race several agents head-to-head on the same task in per-task worktrees with tour-assisted review, free on macOS or Linux: Parallel Code, accepting solo-maintainer risk and a fast-churning surface.
• Do not adopt ClawTeam today; it is dormant with no release since March 2026, and its value is as a reference design for the spawn-worktree-mailbox pattern.
• Want every agent CLI you already run, on every machine you own, in one encrypted native window with phone and desk-device reach: OpenHarness, accepting vendor-operated transport, launch-day-only coverage, and macOS or Linux only.
• Want an agent whose green verdicts arrive as signed, offline-verifiable bundles with a retry ratchet that protects your tests: Swarm Orchestrator, accepting a solo maintainer, four products in one package name, and a Node 24 CLI.
• Want plan-to-PR discipline over the CLI agents you already run, on Windows as well as macOS and Linux: Ivy Tendril, accepting the FSL license, a mid-rewrite codebase silent for a month, and docs domains that do not resolve.

Changes

#

• 2026-08-24 - Created with seven tools and the death and orphan stories carried into the reading section.

• 2026-08-27 - Extended from seven to eight columns with Gas Town and canonicalized the emdash reference.

• 2026-08-30 - Added Paseo, Superset, and Worktrunk columns, reaching eleven, with a funding-spectrum sentence.

• 2026-09-06 - Extended from twelve to thirteen columns with Happy Coder.

• 2026-09-13 - Corrected the Happy Coder and JetBrains Air columns, whose body cells had been transposed since the Air column was added on 2026-09-12.

• 2026-09-13 - Reframed Gas Town’s status to active but cooling (no default-branch commit since 2026-07-23, no release since v1.2.1 in June).

• 2026-09-20 - Moved Gas Town’s status cell to shut down (September 2026, Yegge’s admission per Dan Luu via AINews), matching the note’s death record.

• 2026-09-13 - Corrected the Vibe Kanban orphan wording to no default-branch commit since 2026-04-24, named Omnara Cloud and JetBrains cloud tasks in the cloud-execution prose, and qualified Happy Coder’s star ranking as second among maintained tools.

• 2026-09-16 - Re-verification: recorded Vibe Kanban’s first post-shutdown default-branch commit (2026-09-15), refreshed the Omnara star count to 2,851, and updated the JetBrains Air release to 262.834.41.

• 2026-09-18 - Re-verification: recorded Vibe Kanban’s community commits resuming (eight on the default branch since 2026-09-15, still no release), moved the cmux price cell to the new monthly-only Pro $50 and Max $200 tiers, refreshed the Omnara star count to 2,857, and corrected the stale Air changelog version in the references.

• 2026-09-20 - Re-verification: moved the Vibe Kanban status cell to the community’s 0.1.45 tag, which npm and the release list do not publish yet, and refreshed the Omnara star count to 2,861.

• 2026-09-21 - Extended from fourteen to fifteen columns with AX (Google’s cluster-scale agent fleet orchestrator, added sorted into the first position), refreshed the Happy Coder star count to 23.9k, and moved the Paseo status cell to v0.8 with 0.9 betas.

• 2026-09-22 - Extended from fifteen to sixteen columns with Foremerge (Nick Woodhead’s coordination protocol for parallel agents, added sorted after Emdash), moved the Paseo status cell to v0.9.1, and refreshed the Omnara star count to 2,863.

• 2026-09-24 - Removed the verification preamble line on owner request.

• 2026-09-24 - Refreshed the Vibe Kanban status cell to the GitHub v0.1.45 prerelease (npm still 0.1.44), moved Paseo’s status cell to v0.9.2, and refreshed Foremerge’s star count to 505.

• 2026-09-27 - Extended from sixteen to twenty columns with AutoGen, AutoGPT, LobeHub, and MetaGPT, the multi-agent frameworks and hosted platforms, and updated the prose counts and the reading section to match.

• 2026-09-27 - Extended from twenty to thirty-one columns with Crewplane, GraphCode, Helmor, Lanes, LoopTroop, Multica, oh-my-codex, Open Swarm, Orca, Sandcastle, and The Perfect Orchestrator (a batch of workflow runners, a graph workspace, a sandbox library, a tmux harness, an issue-board platform, a Codex workflow layer, and a mission-control canvas), re-sorted all columns, and added the three-family reading section and the new choosing entries.

• 2026-09-29 - Extended from thirty-one to thirty-nine columns with Agno, Atlas, CrewAI, Dify, Flowise, Mastra, Sim, and Squad (the framework-and-platform wave, a source-control-for-agents desktop app, and a Copilot CLI team harness), re-sorted all columns, moved the Foremerge, GraphCode, Lanes, Multica, Omnara, Open Swarm, Orca, and Paseo status cells and Paseo’s price cell to their 2026-09-29 values, and updated the prose counts and the reading and choosing sections.

• 2026-09-30 - Re-verification: moved the AutoGPT status cell to beta v0.8.2 (2026-09-30), the GraphCode cell to v0.1.77 (2026-09-29) with 133 stars, the Orca cell to v1.4.217 (2026-09-29) with 81.9k stars, the Paseo cell to v0.10.2 (2026-09-30), and the Sim cell to v0.9.6 (2026-09-29) with 29.8k stars, and rewrote the Vibe Kanban status cell and reading prose now that its v0.1.45 prerelease release is no longer listed on GitHub (the tag remains, npm still 0.1.44).

• 2026-10-02 - Re-verification: moved the Agno status cell to v3.1.0 (October 1), LoopTroop to v0.6.0 (October 1), Mastra to @mastra/core 1.74.0 (October 1) with 7.1M npm downloads, Multica to v0.6.1 (October 1), oh-my-codex to v0.21.7 (October 1), Open Swarm to v1.8.0-exp.5 (October 2), Orca to v1.4.218 (September 30), and Sim to v0.9.9 (October 1), refreshed the star and as-of values across the status cells, corrected the stale Foremerge cell count (518 to 520), and fixed GraphCode’s age in the reading prose.

• 2026-10-02 - Extended from thirty-nine to forty columns with OpenRig (the local cross-harness control plane added sorted between Open Swarm and Orca), traced all ten new cells to the new note, added the topology-as-artifact reading paragraph and the self-hosted control-plane choosing entry, and bumped the intro, thesis, and status-row counts to forty.

• 2026-10-03 - Extended from forty to forty-four columns with Agent Swarm (self-hosted MIT lead/worker platform), AgentGrid (closed desktop canvas app), AgentsMesh (BSL-1.1 fleet pod console), and Buzz (Block’s Nostr-relay communication platform), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the four-new-columns reading paragraph and four choosing entries, added the new columns’ references, and bumped the intro, thesis, and status-row counts to forty-four.

• 2026-10-03 - Re-verification: moved the Agno cell to v3.1.1 (October 2), Atlas to 8.8k stars, Foremerge to v0.5.1 (October 3), Multica to 51.9k stars, OpenRig to 4.5k stars, Orca to v1.4.219 (October 2) with 84k stars, Paseo to v0.10.3 (October 2), Sandcastle to active again with maintenance commits resumed October 2, Sim to v0.9.12 (October 3), Squad to v1.0.0 (October 3) and no longer alpha, and updated the Sandcastle reading and choosing prose to match.

• 2026-10-04 - Re-verification: moved the cmux cloud-execution cell to the new cap of 5 cloud VMs (the vendor restructured cloud packaging the same day at unchanged dollar tiers), refreshed the Agent Swarm, Atlas, Crewplane, Dify, Foremerge, GraphCode, LoopTroop, Open Swarm, and OpenRig status cells to their 2026-10-04 values, and updated the cmux pricing reference note.

• 2026-10-05 - Re-verification: moved the Agent Swarm status cell to v1.163.0 (October 5), AgentGrid to v2.9.3 (October 4), GraphCode to v0.1.78 (October 4) as a stable release, OpenRig to v0.6.5 (October 4) with 5.0k stars, Orca to v1.4.220 (October 4), Sim to v0.9.13 (October 4), and Squad to v1.0.1 (October 4), and refreshed the Agno, Atlas, AutoGPT, Buzz, CrewAI, Dify, Happy Coder, Mastra, Multica, and oh-my-codex counts to their 2026-10-05 values.

• 2026-10-06 - Extended from forty-four to forty-nine columns with Agent Manager (the nine-CLI tmux TUI), LangGraph (the graph runtime), Raven (the DAG-planning host agent), Ruflo (the claude-flow-renamed meta-harness), and Scion (Google’s container hypervisor), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the five-new-columns reading paragraph, five choosing entries, and the new columns’ references, and bumped the intro, thesis, and status-row counts to forty-nine.

• 2026-10-06 - Extended from forty-nine to fifty-two columns with Agent Orchestrator (the Apache-2.0 kanban workspace at 12.8k stars under its fourth organization name), 1Code (the open agent client dormant since March), and Bernstein (the deterministic-scheduler governance layer), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the trio reading paragraph, three choosing entries, and the new columns’ references, moved thirteen status cells to their 2026-10-06 values and the dmux cell to quiet (no push since 2026-08-16), and bumped the intro, thesis, and status-row counts to fifty-two.

• 2026-10-06 - Extended from fifty-two to fifty-nine columns with Yao (Infinite Wisdom Software’s cross-device agent workspace on the five-year Yao Engine), inserted after Worktrunk in sorted position with all ten cells traced to the new note, added the inherited-stars reading paragraph, one choosing entry, and the Yao references, and bumped the intro, thesis, and status-row counts to fifty-nine.

• 2026-10-07 - Extended from fifty-three to fifty-nine columns with AG2 (AutoGen’s community fork), Agent Teams (AGPL boss-and-team desktop), Agentlas OS (hub with a per-task temporary orchestrator), CAMEL (the 2023 role-playing research lineage), hcom (terminal-to-terminal agent messaging fabric), and PraisonAI (low-code self-reflection framework), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, and moved the 1Code status cell to dead (archived by its owner 2026-07-07), matching its note.

• 2026-10-07 - Re-verification: moved the Agent Manager, Agent Swarm, AgentGrid, Buzz, LangGraph, Lanes, OpenRig, Orca, and Ruflo status cells to their 2026-10-07 versions.

• 2026-10-07 - Extended from fifty-nine to sixty-one columns with Archon (Cole Medin’s MIT YAML workflow engine over the Claude Code and Codex SDKs with per-run worktrees) and T3 Code (Ping.gg’s MIT control surface with native iOS, Android, web, and Electron clients driving six harnesses), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the two-new-columns reading paragraph, two choosing entries, and the new columns’ references, and bumped the intro, thesis, and status-row counts to sixty-one.

• 2026-10-08 - Extended from sixty-one to sixty-two columns with Hive (Aden’s Apache-2.0 colony runtime whose Queen spawns worker clones over a shared ledger), inserted in case-insensitive alphabetical position with all ten cells traced to the new note, added a reading paragraph, one choosing entry, and the new column’s references, and bumped the intro, thesis, and status-row counts to sixty-two.

• 2026-10-08 - Re-verification: moved the Agent Orchestrator, Agent Swarm, Agent Teams, AgentGrid, Agentlas OS, AutoGPT, Buzz, CrewAI, Dify, LobeHub, Mastra, Multica, OpenRig, Paseo, PraisonAI, Ruflo, Superset, and T3 Code status cells to their 2026-10-08 values, and re-sequenced the three 2026-10-06 bullets into chronological position after a script had appended them after the 2026-10-07 ones.

• 2026-10-09 - Re-verification: moved the Agent Manager, Agent Orchestrator, Agent Swarm, Agentlas OS, Agno, CrewAI, LangGraph, Open Swarm, OpenRig, Orca, PraisonAI, Ruflo, and Sim status cells to their 2026-10-09 values, including the LangGraph npm-downloads figure now at 3.73M weekly installs (down from 4.55M) with the reading prose and references note updated to match.

• 2026-10-08 - Recorded here because the append-only rule forbids rewriting it: the 2026-10-06 Yao bullet’s from-count reads fifty-nine where fifty-three is the count that chains with the surrounding bullets (one column added that day, not seven).

• 2026-10-10 - Extended from sixty-two to sixty-three columns with Symphony (OpenAI’s Apache-2.0 spec-first scheduler turning tracker issues into isolated autonomous implementation runs, the queue-flagged carve-out from the mission-control-tail rejection), inserted in case-insensitive alphabetical position after Superset with all ten cells traced to the new note, added a reading paragraph, one choosing entry, and the new column’s references, and bumped the intro, thesis, and status-row counts to sixty-three.

• 2026-10-10 - Re-verification: moved the Agent Orchestrator, Agent Teams, Agentlas OS, Buzz, CrewAI, OpenRig, Orca, Paseo, and Ruflo status cells to their 2026-10-10 values, refreshed the Atlas, LangGraph (3.77M weekly npm installs, the reading prose and references note updated to match), Multica, and T3 Code cells, and corrected the Bernstein cell from 1.4k to 1.2k stars, which the 2026-10-06 verification had recorded against an API count now 200 lower.

• 2026-10-10 - Extended from sixty-three to sixty-six columns with herdr (the Rust terminal-runtime daemon whose agent automation API makes it the category’s first column beneath the harnesses, admitted past the mission-control family on the scanner’s recommendation), AgentBox (Marco D’Alia’s MIT box teleporter with sub-second checkpoints across Docker and five clouds, the sandboxing scan’s orchestration-fit flag), and EvoAgentX (the self-evolving multi-agent workflow framework the automated-research scan referred here), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the three-new-columns reading paragraph, three choosing entries, and the new columns’ references, and bumped the intro, thesis, and status-row counts to sixty-six.

• 2026-10-12 - Extended from sixty-six to sixty-nine columns with Jean (the Coolify team’s Apache-2.0 dev environment putting local and jean-server projects in one sidebar), Gas City (the MIT orchestration-builder SDK extracted from the shut-down Gas Town, the category’s first build-your-own-orchestrator toolkit), and Orkas (the MIT commander-and-specialist desktop whose nine specialists ship installed), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, moved the Open Swarm column to the pivot record (proprietary Open Swarm Inc. since the privacy policy’s 2026-10-03 effective date, repo a single-commit signed-builds shell, v1.8.0 build 2026-10-11, license, price, cloud, and status cells rewritten), added the three-new-columns reading paragraph, three choosing entries, and the new columns’ references, and bumped the intro, thesis, and status-row counts to sixty-nine.

• 2026-10-12 - Re-verification: fixed the intro count that still read sixty-six after the morning’s bump to sixty-nine, corrected the Gas City status cell (v1.5.0 stable shipped 2026-10-05, so the earlier no-stable-release read was wrong) and its choosing entry, moved the AG2 (v1.2.0), GraphCode (v0.1.80), oh-my-codex (v0.21.9), OpenRig (v0.6.10, 7.0k stars), Paseo (20.6k), Ruflo (v3.56.7), and Sim (v0.9.18) status cells and the Orca (90.1k), AutoGen (MAF 14.1k), and LangGraph (3.80M npm/week) counts to their 2026-10-12 values, and widened the JetBrains Air harness cell to the homepage’s six named agents plus ACP.

• 2026-10-13 - Re-verification: moved the Yao status cell to its first stable major (v1.0.0, 2026-10-13, with the choosing entry’s pre-1.0 wording retired) and corrected the Hive cell and reading prose after its trunk woke with a README documentation merge on 2026-10-12, moved the Ruflo (v3.56.9), oh-my-codex (v0.21.11), Sim (v0.9.19), Agent Swarm (v1.171.1), Agno (v3.1.3), hcom (v0.7.29), and LangGraph (v1.2.15) cells to their newest release trains, and refreshed the Orca (90.9k), herdr (43.7k), Multica (52.6k), OpenRig (7.3k), Paseo (20.7k), LobeHub (83.2k), Gas City (1,368), and Orkas (2,177) counts to their 2026-10-13 values.

• 2026-10-19 - Re-verification: moved the Agent Swarm (v2.4.0), Crewplane (v0.3.8), oh-my-codex (v0.21.19), Orca (v1.4.230), Paseo (v0.12.0), Hive (v1.0.10), Ruflo (v3.57.0), Sim (v0.9.26), OpenRig (v0.6.11), Agentlas OS (v1.2.66), Jean (v1.0.12), LobeHub (v2.2.20), and Orkas (v2026.10.17) status cells to their newest release trains, star figures untouched.

• 2026-10-20 - Re-verification: moved the Agno (v3.1.6), Agent Orchestrator (v0.14.3), Agent Swarm (v2.4.2), Agentlas OS (v1.2.67), GraphCode (v0.1.81), oh-my-codex (v0.21.25), OpenRig (v0.6.13), Orca (v1.4.232), PraisonAI (v4.7.16), Ruflo (v3.58.0), Sim (v0.9.28), and Superset (v1.39.0) status cells to their newest release trains, star figures untouched.

• 2026-10-21 - Extended from sixty-nine to seventy-three columns with Frink (benord-labs’ MIT desktop app pairing four-pane parallel agents with schedule-, webhook-, and event-triggered Flows), gwq (d-kuro’s Apache-2.0 worktree-manager CLI with a global cross-repository view, the Worktrunk seat), YYLO (JUNO AI’s MIT bounded-task orchestrator with a Git-native ledger and protected merges), and apowerb (thaink²’s Apache-2.0 open-core agent platform on Google ADK and LiteLLM), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the four-new-columns reading paragraph, four choosing entries, and the new columns’ references, and bumped the intro, thesis, and status-row counts to seventy-three.

• 2026-10-23 - Extended from seventy-three to seventy-five columns with ClawTeam (HKUDS’s dormant MIT swarm CLI whose leader agent spawns worktree-and-tmux workers coordinating through CLI mailboxes, admitted past the thin-tail shelf on scale and provenance) and Parallel Code (Johannes Millan’s MIT Electron desktop racing coding agents in per-task worktrees, the worktree-CLI seat), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the two-new-columns reading paragraph, two choosing entries, and the new columns’ references, and bumped the intro, thesis, and status-row counts to seventy-five.

• 2026-10-24 - Extended from seventy-five to seventy-seven columns with OpenHarness (Autonomous’s MIT native command center running fourteen existing CLIs in tmux panes and worktrees across end-to-end-encrypted linked machines, admitted on the standalone-ADE precedent with the launch-echo coverage and vendor-only benchmarks recorded as cautions) and Swarm Orchestrator (the solo-authored evidence-carrying coding agent with worktree workers and offline-verifiable bundles, admitted past the small-member floor with the missing-footprint signal stated), inserted in case-insensitive alphabetical position with all ten cells per column traced to the new notes, added the two-new-columns reading paragraph, two choosing entries, and the new columns’ references, and bumped the intro, thesis, and status-row counts to seventy-seven.

• 2026-10-24 - Extended from seventy-seven to seventy-eight columns with Ivy Tendril (Ivy Interactive’s source-available plan-to-pull-request software factory driving any CLI agent in parallel worktrees behind human review gates, completed same-run from the cross-category scan’s admission with the mid-rewrite status lead and the broken-docs-domains caution), inserted in case-insensitive alphabetical position between Hive and Jean with all ten cells traced to the new note, added the reading paragraph, one choosing entry, and the new column’s references, and bumped the intro, thesis, and status-row counts to seventy-eight.

See also

#

• Harness Feature Matrix - the same treatment for the harness layer these tools drive
• Surface Feature Matrix - the same treatment for editors and environments
• Agentic Coding Tools Landscape - the map these columns come from
• Managing Many Concurrent LLM Agent Sessions - the supervision problem the whole category answers
• OpenChamber - scheduled sessions, the non-interactive complement to parallel dispatch

References

#

• https://github.com/smtg-ai/claude-squad - profiles, worktrees, packaging for the Claude Squad column
• https://github.com/google/ax - repository, primitives, license, and status for the AX column
• https://agentexecutor.io - the Task, Workspace, Gateway, and Model model for the AX column
• https://cmux.com/pricing - tiers, the 5-VM cloud cap (50 before 2026-10-04), CodeRouter removal for the cmux column
• https://conductor.build/pricing/ - tiers, sandbox specs, local versus cloud privacy for the Conductor column
• https://github.com/stravu/crystal - deprecation notice and feature history for the Crystal column
• https://github.com/standardagents/dmux - supported agent list, hooks, local-only scope for the dmux column
• https://emdash.com/ - agent support, SSH model, downloads for the Emdash column
• https://github.com/BloopAI/vibe-kanban - sunset banner and feature list for the Vibe Kanban column
• https://github.com/gastownhall/gastown - architecture, runtimes, Refinery, Docker setup for the Gas Town column
• https://github.com/slopus/happy - repository, stars, license, and monorepo layout for the Happy Coder column
• https://paseo.sh/alternatives/happy-coder - the provider, worktree, and platform limits for the Happy Coder column
• https://news.ycombinator.com/item?id=44904039 - launch thread and donation-IAP context for the Happy Coder pricing cell
• https://maggieappleton.com/gastown - the field analysis grounding the Gas Town status and cost cells
• https://github.com/omnara-ai/omnara - control plane, license, and stars for the Omnara column
• https://www.vibekanban.com/blog/shutdown - shutdown date, service removal, refunds for the status row
• https://github.com/getpaseo/paseo - platforms, agent catalog, relay, and license for the Paseo column
• https://github.com/superset-sh/superset - worktree model, presets, ELv2 license, and funding for the Superset column
• https://github.com/max-sixty/worktrunk - commands, hooks, license, and release cadence for the Worktrunk column
• https://air.dev/ - supported agents, isolation options, and pricing FAQ for the JetBrains Air column
• https://air.dev/changelog - current version 262.834.41 and platform dates for the JetBrains Air status cell
• https://www.jetbrains.com/help/air/execution-environments.html - the worktree, Docker, and cloud isolation model for the JetBrains Air isolation cells
• https://www.jetbrains.com/help/air/cloud-tasks.html - the JetBrains-managed cloud environments for the JetBrains Air cloud cell
• https://github.com/stablyai/orca - repository, worktrees, 27-agent list, MIT license, and stars for the Orca column
• https://www.onorca.dev/docs - worktrees, review, remote, and the “not a model” scope for the Orca cells
• https://github.com/mattpocock/sandcastle - sandbox providers, branch strategy, hooks, and license for the Sandcastle column
• https://registry.npmjs.org/@ai-hero%2Fsandcastle - Sandcastle package version and license
• https://lanes.sh/pricing - tiers, seats, and Compute billing for the Lanes price cell
• https://github.com/lanes-sh/app - board, sessions, worktrees, and the undeclared license for the Lanes column
• https://github.com/crewplaneai/crewplane - Markdown workflows, resumable runs, and license for the Crewplane column
• https://helmor.ai/ - local-first workbench, agents, and ship actions for the Helmor column
• https://github.com/dohooo/helmor - worktrees, one-click PR and merge, CLI and MCP, and the cooling commit cadence for Helmor
• https://github.com/openswarm-ai/openswarm - canvas, approvals, worktrees, and the AGPL-3.0 LICENSE for the Open Swarm column
• https://www.looptroop.ovh/docs/ - council planning, beads, Ralph loops, and alpha status for the LoopTroop column
• https://graphcode.app/ - loop types, graph edges, and the FSL license for the GraphCode column
• https://github.com/daman8271/the-perfect-orchestrator - lead-and-worker harness, adversarial verification, and license for the column
• https://github.com/multica-ai/multica - issue assignment, runtimes, license metadata, and stars for the Multica column
• https://www.promptquorum.com/power-local-llm/multica-review - the star-count caveat and the custom-license analysis for the Multica cells
• https://github.com/Yeachan-Heo/oh-my-codex - skills, worktree teams, MCP servers, and license for the oh-my-codex column
• https://github.com/mvschwarz/openrig - repository, Apache-2.0 license, stars, and the setup footprint that writes provider trust and hooks for the OpenRig column
• https://github.com/desplega-ai/agent-swarm - repository, MIT license, lead/worker architecture, and stars for the Agent Swarm column
• https://docs.agent-swarm.dev - intake surfaces, Docker worker model, workflows, and the harness list for the Agent Swarm column
• https://agentgrid.sh/ - the canvas, panes, master/worker MCP delegation, and harness list for the AgentGrid column
• https://agentgrid.sh/pricing - the Free and Pro tiers behind the AgentGrid price cell (fetched 2026-10-02; HTTP 500 on re-fetch 2026-10-03)
• https://github.com/AgentsMesh/AgentsMesh - repository, BSL-1.1 license, AgentPod architecture, and stars for the AgentsMesh column
• https://github.com/block/buzz - repository, Apache-2.0 license, surfaces, and stars for the Buzz column
• https://openrig.dev/what-is - the rig, pod, and seat model and the local control-plane architecture for the OpenRig cells
• https://openrig.dev/docs/getting-started - prerequisites, starters, and the cross-harness team model for the OpenRig column
• https://news.ycombinator.com/item?id=48241066 - the isolation question that grounds the OpenRig worktree and sandbox cells
• https://news.ycombinator.com/item?id=47772935 - the supervision concern raised on the OpenRig launch thread
• https://api.github.com/repos/ruvnet/ruflo - stars, MIT license, and push date for the Ruflo column
• https://github.com/ruvnet/claude-flow - the 301 redirect to ruflo that grounds the Ruflo provenance cell
• https://cognitum.one/agentic-engineering - the company page behind the Ruflo banner
• https://news.ycombinator.com/item?id=49051072 - the three-point Show HN grounding the Ruflo adoption caveat
• https://docs.langchain.com/oss/python/langgraph/ - durable execution, interrupts, persistence, and the LangSmith split for the LangGraph column
• https://api.npmjs.org/downloads/point/last-week/@langchain/langgraph - the weekly npm installs figure for the LangGraph status cell (3.80M as of 2026-10-12)
• https://news.ycombinator.com/item?id=43468435 - qodo’s coding-agent thread for the LangGraph column
• https://googlecloudplatform.github.io/scion/overview/ - the run modes and Hub/broker split for the Scion column
• https://www.infoq.com/news/2026/04/google-agent-testbed-scion/ - the hypervisor framing and partial Codex/OpenCode support for the Scion column
• https://news.ycombinator.com/item?id=47675213 - the 230-point Scion launch thread
• https://raven.evermind.ai - the RSI positioning grounding the Raven column
• https://news.ycombinator.com/item?id=49890647 - the AI-comments warning grounding the Raven caution cells
• https://agent-manager.dev - the FAQ, license history, and nine-CLI list for the Agent Manager column
• https://news.ycombinator.com/item?id=49107749 - the distinguishability debate grounding the Agent Manager cells
• https://github.com/OrchestratorInc/agent-orchestrator - repository, harness count, license, and stars for the Agent Orchestrator column
• https://docs.orchestrator.inc/guides/cloud - the account-gated cloud sessions behind the Agent Orchestrator cloud cells
• https://starlog.is/articles/ai-agents/untrivial-ai-agent-orchestrator - the architecture review grounding the Agent Orchestrator kind and review cells
• https://github.com/21st-dev/1code - repository, feature list, and the March 2026 last push for the 1Code column
• https://news.ycombinator.com/item?id=46637723 - the launch thread and $20/month hosted tier for the 1Code price cells
• https://github.com/sipyourdrink-ltd/bernstein - repository, deterministic-scheduler positioning, and license for the Bernstein column
• https://docs.bernstein.run/en/latest/ - the policy, worktree, and audit model for the Bernstein cells
• https://www.augmentcode.com/tools/open-source-agent-orchestrators - the hands-on roundup grounding the Bernstein verification cell
• https://api.github.com/repos/YaoApp/yao - repository, counts, and topics for the Yao column
• https://api.github.com/repos/YaoApp/yao/license - the modified Apache-2.0 terms grounding the Yao license and price cells
• https://yaoagents.com - the Yao Agents, Tai Link, and Yao Engine product split grounding the Yao kind and platforms cells
• https://yaoagents.com/blog/en-us/2026/release/deepseek-harness - the first-party DeepSeek Harness integration grounding the Yao harness cell
• https://hn.algolia.com/api/v1/items/30699919 - the 2-point 2022 low-code launch grounding the Yao status and reading prose
• https://github.com/coleam00/Archon - the Archon column: repository, workflow model, and license (as of 2026-10-07)
• https://archon.diy/docs/ - the Archon column: the workflow engine definition, providers, and trigger surfaces
• https://github.com/pingdotgg/t3code - the T3 Code column: repository, harness list, and license (as of 2026-10-07)
• https://t3.codes - the T3 Code column: the four client surfaces, bring-your-own-subscription rule, and PR flow
• https://github.com/aden-hive/hive - the Hive column: repository, colony model, Sentinel, and license (as of 2026-10-08)
• https://docs.adenhq.com/ - the Hive column: the framework docs, quickstart, and Aden Cloud deployment
• https://www.ycombinator.com/companies/aden - the Hive column: Aden’s Winter 2020 batch, five-person team, and Acho history
• https://github.com/openai/symphony - the Symphony column: repository, preview warning, spec-first positioning, and license (as of 2026-10-10)
• https://raw.githubusercontent.com/openai/symphony/main/SPEC.md - the Symphony column: the Draft v1 spec’s goals, non-goals, and implementation-defined policy
• https://raw.githubusercontent.com/openai/symphony/main/elixir/README.md - the Symphony column: Codex App Server mode, the five tracker adapters, and host-side credential handling
• https://api.github.com/repos/herdrdev/herdr - the herdr column: repository, license, and counts (as of 2026-10-10)
• https://herdr.dev/docs/session-state/ - the herdr column: what survives a restart and the 18-CLI native session restore table
• https://herdr.dev/docs/agent-automation/ - the herdr column: the automation primitives and the 24 supported agent kinds
• https://herdr.dev/blog/herdr-is-joining-y-combinator/ - the herdr column: YC F26, the AGPL-to-Apache switch, and the plugin count
• https://herdr.dev/blog/herdr-raised-a-seed/ - the herdr column: the $6M Bessemer-led seed
• https://news.ycombinator.com/item?id=48756578 - the herdr column: the 404-point thread and the tmux critique
• https://api.github.com/repos/madarco/agentbox - the AgentBox column: repository, license, and counts (as of 2026-10-10)
• https://raw.githubusercontent.com/madarco/agentbox/main/README.md - the AgentBox column: the box model, provider table, and checkpoint claims
• https://registry.npmjs.org/@madarco/agentbox/latest - the AgentBox column: npm 0.33.0 and the sandbox provider modules
• https://news.ycombinator.com/item?id=49728120 - the AgentBox column: the author’s Show HN description of the credential relay
• https://api.github.com/repos/ANative-Lab/EvoAgentX - the EvoAgentX column: repository, dates, and license detection (as of 2026-10-10)
• https://arxiv.org/abs/2507.03616 - the EvoAgentX column: the framework paper’s architecture and benchmark claims
• https://arxiv.org/abs/2508.07407 - the EvoAgentX column: the team’s self-evolving-agents survey
• https://hn.algolia.com/api/v1/search?query=EvoAgentX&hitsPerPage=5 - the EvoAgentX column: the self-posted-only HN footprint
• https://github.com/coollabsio/jean - the Jean column: repository, counts, Apache-2.0 license, and the nine-backend list (as of 2026-10-12)
• https://jean.build - the Jean column: the local-and-remote sidebar model, the free-forever positioning, and the macOS-tested caveat
• https://github.com/gastownhall/gascity - the Gas City column: repository, counts, MIT license, and the runtime-provider list (as of 2026-10-12)
• https://docs.gascityhall.com - the Gas City column: the six primitives and the software-factory positioning
• https://docs.gascity.com/getting-started/coming-from-gastown - the Gas City column: the role-to-configuration mapping tables grounding the extraction claim
• https://github.com/Orkas-AI/Orkas - the Orkas column: repository, counts, MIT license, and the shipped-specialist table (as of 2026-10-12)
• https://orkas.ai/pricing/ - the Orkas column: the credit tiers and the Stripe sandbox checkout notice
• https://api.github.com/repos/benord-labs/frink - the Frink column: repository, counts, license, and dates (as of 2026-10-21)
• https://raw.githubusercontent.com/benord-labs/frink/main/user-docs/flows.md - the Frink column: flow anatomy, triggers, and the Claude/OpenAI rule
• https://api.github.com/repos/d-kuro/gwq - the gwq column: repository, counts, and dates (as of 2026-10-21)
• https://raw.githubusercontent.com/d-kuro/gwq/main/docs/release-notes/v0.1.2.md - the gwq column: recursive copy_files and the global-discovery speedup
• https://api.github.com/repos/apowerb/apowerb - the apowerb column: repository, counts, and dates (as of 2026-10-21)
• https://docs.apowerb.com/concepts/editions - the apowerb column: the open-core boundary and the absent commercial bricks
• https://api.github.com/repos/yylo-dev/yylo - the YYLO column: repository, counts, and dates (as of 2026-10-21)
• https://raw.githubusercontent.com/yylo-dev/yylo-ledger/main/README.md - the YYLO column: the Git-native Ledger and the exact-version pin
• https://api.github.com/repos/HKUDS/ClawTeam - the ClawTeam column: repository, counts, and dates (as of 2026-10-23)
• https://raw.githubusercontent.com/HKUDS/ClawTeam/main/README.md - the ClawTeam column: the spawn, worktree, mailbox, and board model
• https://hn.algolia.com/api/v1/search?query=ClawTeam&tags=story - the ClawTeam column: the absent HN footprint grounding the caution cells
• https://api.github.com/repos/johannesjo/parallel-code - the Parallel Code column: repository, counts, and dates (as of 2026-10-23)
• https://raw.githubusercontent.com/johannesjo/parallel-code/main/README.md - the Parallel Code column: worktree automation, tours, arena, and harness list
• https://hn.algolia.com/api/v1/items/47115578 - the Parallel Code column: the two-point Show HN grounding the adoption cells
• https://api.github.com/repos/autonomous-ai/openharness - the OpenHarness column: repository, counts, and dates (as of 2026-10-24)
• https://raw.githubusercontent.com/autonomous-ai/openharness/main/README.md - the OpenHarness column: engines, transport, store, and device
• https://raw.githubusercontent.com/autonomous-ai/openharness/main/docs/engines.md - the OpenHarness column: the fourteen-engine table
• https://www.autonomous.ai/harness - the OpenHarness column: the vendor page and the hardware prices
• https://reveneau.com/ainews/openharness-autonomous-command-center-coding-agents-mit-1266-stars - the OpenHarness column: launch facts and the vendor-measurement caveat
• https://api.github.com/repos/moonrunnerkc/swarm-orchestrator - the Swarm Orchestrator column: repository, counts, and dates (as of 2026-10-24)
• https://raw.githubusercontent.com/moonrunnerkc/swarm-orchestrator/main/README.md - the Swarm Orchestrator column: the loop, parallel mode, and evidence model
• https://raw.githubusercontent.com/moonrunnerkc/swarm-orchestrator/v13-main/docs/claims.md - the Swarm Orchestrator column: the claim-to-artifact table and forbidden claims
• https://api.github.com/repos/Ivy-Interactive/Ivy-Tendril - the Ivy Tendril column: repository, counts, and dates (as of 2026-10-24)
• https://raw.githubusercontent.com/Ivy-Interactive/Ivy-Tendril/development/README.md - the Ivy Tendril column: the feature table, agent list, and skills catalog
• https://api.github.com/repos/Ivy-Interactive/Ivy-Tendril/releases?per_page=5 - the Ivy Tendril column: the release train ending at v1.2.4 (2026-09-14)
• https://raw.githubusercontent.com/Ivy-Interactive/Ivy-Tendril/development/LICENSE - the Ivy Tendril column: the FSL-1.1-ALv2 terms
• https://api.github.com/repos/Ivy-Interactive/Ivy-Tendril-V2 - the Ivy Tendril column: the V2 monorepo’s dates and WIP positioning
• https://raw.githubusercontent.com/Ivy-Interactive/Ivy-Tendril-V2/development/README.md - the Ivy Tendril column: the Tauri, React, and Rust layout and the delete-when-done notice
• https://ivy.app - the Ivy Tendril column: the plan-draft-review lifecycle and the local-first claims
• https://ivy.app/pricing - the Ivy Tendril column: the Free, Pro $59, and Enterprise tiers
• https://ivy.app/llms.txt - the Ivy Tendril column: the company facts and the use-and-do-not-use guidance
• https://hn.algolia.com/api/v1/search?query=%22Ivy+Tendril%22&hitsPerPage=10 - the Ivy Tendril column: the zero-story HN footprint