run-recording on tomrochette.com

Agent-Causality

tom@tomrochette.com (Tom Rochette) — Wed, 28 Oct 2026 00:00:00 +0000

Agent-Causality is an AGPL-3.0 Python SDK and CLI that records OpenCode and LangGraph agent runs as causal event graphs with declared parents, slices a failing event to the recorded events it structurally depends on, and replays registered decisions with changed inputs to test which ones flip the outcome.

Agent-Causality’s bet is that causality belongs in the capture, not the post-mortem: every event carries the parents its recorder declared, so slices, provenance, and counterfactual replay are queries over facts the run wrote down, and the project’s own benchmark reports where that stops working.

What it is
#

Two packages are required, and the README says so plainly: agent-causality on PyPI (the SDK, CLI, an ingest receiver for OpenCode events, and the Visual DAG Explorer web UI) and @psionic-labs/opencode-agent-causality on npm (an OpenCode plugin, telemetry capture only, forwarding the global event stream plus tool, permission, and message hooks). Events persist to one SQLite file per run, with secrets redacted before persistence; capture is fail-open and never interrupts the agent. LangGraph runs record through a callback handler that converts parent_run_id containment into explicit causal edges, dropping unknown parents rather than inventing them. The Explorer renders the causal graph and a per-agent timeline and ships analysis tabs (chain, provenance, interaction, evidence, metrics), no API key needed; an optional explain extra adds LLM-grounded analysis through OpenRouter, kept in its own labeled block. Made by Vasant Kumar Mogia (Psionics) under AGPL-3.0, with an in-repository research paper on declared dependency tracing and counterfactual sensitivity analysis.

Status
#

Young and active but essentially unadopted: 6 stars, 2 forks, 3 open issues, created 2026-08-15, pushed 2026-10-28, releases through v0.2.1 on 2026-10-22, as of 2026-10-28. The registries confirm the thinness: PyPI serves 0.2.1 (the README names 0.3.0) and the OpenCode plugin 0.1.6, with 154 PyPI downloads in the week to 2026-10-28 and 243 npm downloads in the month to 2026-10-26.

The critical source is the project’s own benchmark, and it is unusually frank: live OpenCode runs captured at micro/macro-average 1.0, but a cause recall of 0.3 over explicit-evidence edges and a joint ancestry of 0.0, with the README conceding that genuine causal interaction is unsupported from live traces. I found no independent discussion: the Hacker News story search returns zero hits, and the README’s evidence is its own benchmark results. A solo research-grade tool with a paper, frankly negative self-benchmarks, and no community verdict yet, so read the benchmark results before believing the causal claims.

Strengths
#

Declared, not inferred, causal edges: every parent link is recorded at capture time by the recorder that saw the dependency happen.
The self-benchmark names its failures (cause recall 0.3, joint ancestry 0.0 on live traces), which is more falsifiable than the category’s usual claims.
Both surfaces a debugging session needs are included: a queryable SQLite graph and a no-key Explorer with graph, timeline, and metrics views.
Counterfactual replay re-runs registered merge decisions with changed inputs and reports which flips change the outcome.
Capture never blocks the agent, and secrets are redacted before anything persists.

Cautions
#

6 stars, 2 forks, and zero independent coverage: every claim here is the project’s own evidence.
The tool’s own live-trace results limit it to structural slices over explicit edges, not true causal interaction, which is the whole promise.
AGPL-3.0 complicates embedding in commercial tooling, and the two-package install means either half alone is incomplete.
Version drift: the README names agent-causality 0.3.0 while PyPI serves 0.2.1, an early-project smell to watch.
One maintainer, pre-1.0, and a schema surface (SQLite event graphs) with no compatibility guarantees yet.

Pricing
#

Free and open source under AGPL-3.0; no paid tier is published, and everything runs locally.

Compared to
#

NovaFabric: the replay-and-evidence sibling from the same source list; NovaFabric preserves and diffs whole runs for audit, Agent-Causality slices recorded runs for cause.
agentsview: the archive that answers history and cost questions; Agent-Causality answers only why this run failed, from a capture it had to be present for.
ClawTrace: the hosted tracing platform with an AI analyst; ClawTrace sees live OpenClaw payloads in the cloud, Agent-Causality keeps declared causal graphs in a local SQLite file.

Bottom line
#

Recommended for engineers running branching LangGraph or OpenCode systems who want failure diagnosis grounded in capture-time facts, and who will read the benchmark before trusting the slice. Not for cost reporting, transcript search, live dashboards, or anyone who needs an independently verified dependency.

Changes
#

2026-10-28 - Created from the coordinator’s entrant handoff, admitted on the declared-causal-parent capture mechanism, the self-critical live-trace benchmark with its named negative results, and nine fetched sources.

References
#

https://api.github.com/repos/Psionic-labs/Agent-Causality - 6 stars, 2 forks, AGPL-3.0, created 2026-08-15, pushed 2026-10-28, as of 2026-10-28
https://raw.githubusercontent.com/Psionic-labs/Agent-Causality/main/README.md - the two-package contract, capture hooks, SQLite storage, Explorer views, benchmark summary, and the paper link
https://api.github.com/repos/Psionic-labs/Agent-Causality/releases - v0.2.1 (2026-10-22), v0.2.0 (2026-10-18), v0.1.4 (2026-10-14)
https://pypi.org/pypi/agent-causality/json - PyPI package, version 0.2.1, Python 3.11 or newer
https://registry.npmjs.org/@psionic-labs/opencode-agent-causality - the OpenCode plugin, latest 0.1.6 (2026-10-22)
https://pypistats.org/api/packages/agent-causality/recent - 154 downloads in the week to 2026-10-28, behind the adoption claim
https://api.npmjs.org/downloads/point/last-month/@psionic-labs/opencode-agent-causality - 243 downloads in the month to 2026-10-26
https://psionic-labs.github.io/Agent-Causality/opencode.html - the OpenCode integration guide
https://raw.githubusercontent.com/Psionic-labs/Agent-Causality/main/docs/benchmarks.md - the full benchmark results behind the negative-results claim
https://hn.algolia.com/api/v1/search?query=Agent-Causality&tags=story - zero stories, the empty independent footprint this note states

NovaFabric

tom@tomrochette.com (Tom Rochette) — Wed, 28 Oct 2026 00:00:00 +0000

NovaFabric is an Apache-2.0 Python CLI that records any command’s AI-agent run as a portable, secret-scanned Run Capsule you can replay in five modes, structurally diff against another run, trace for lineage, and export as a signed evidence bundle an auditor can verify without NovaFabric installed.

Where the rest of this category indexes what coding agents already wrote to disk, NovaFabric wraps the run itself, so the evidence is a folder you own, sealed with your key, verifiable offline, and diffable run against run.

What it is
#

nova capture <command> injects a loader that hooks the OpenAI, Anthropic, and MCP SDKs plus the httpx, requests, aiohttp, and urllib3 clients, recording each call with OpenTelemetry GenAI fields; non-Python clients (the README names Claude Code, Cursor, and Node or Go agents) ride the api-proxy and mcp-proxy, and an SDK decorator covers in-process capture. Capsules are written on success and on failure, carry a hash-chained redaction proof from an 18-rule secret scan (the README is explicit that this is secret-scanned, not guaranteed secret-free), and stay plain directories on your filesystem. Replay has five named modes: forensic (read-only), mocked (recorded replies served back for CI re-execution), semantic, exact, and experimental intervention; the README does not claim exact replay of remote LLM calls. nova diff aligns two runs call by call and fails a build on behavioral change with --assert-no-regressions; lineage queries export OpenLineage and W3C PROV; sealing is opt-in DSSE with a Merkle log and RFC 3161 timestamping; the exported Evidence Bundle verifies with sha256sum and an Ed25519 verifier and no NovaFabric runtime. Extras add a local dashboard, aggregation queries, an experimental multi-user server with OIDC and RBAC, compliance exports, and Sigstore; the toolkit runs from a laptop to Docker, Kubernetes, and SLURM, online or air-gapped. Made by Mohsen Seyedkazemi Ardebili, an independent consultant, under Apache-2.0, with an arXiv paper (2609.12582) on tamper-evident, replayable evidence for agent runs.

Status
#

Beta and shipping at a clip its adoption has not caught up with: 7 stars, 1 fork, 6 open issues, created 2026-06-25, pushed 2026-10-26, releases through v0.108.0 on 2026-10-25, as of 2026-10-28; PyPI serves 0.108.0 with 734 downloads in the trailing month to 2026-10-28.

The status section labels server mode, the dashboard, and the cluster collector experimental, and the capsule and evidence-bundle formats stay unfrozen until the v1.0 schema freeze. The project’s own comparison table names where it loses (no hosted dashboard, no turnkey live analytics), and an arXiv paper grounds the tamper-evidence design. I found no Hacker News story and no third-party review. A solo project whose documentation discipline outruns its adoption by three orders of magnitude, whose product is an evidence format that is not yet frozen.

Strengths
#

Capture without code changes reaches non-Python agents through the proxies, which most SDK-only tracers cannot.
A crash is captured evidence, not lost state: capsules are written on failure too.
The diff gate turns replay infrastructure into a CI check on agent behavior change.
Evidence Bundles verify with standard tools and no NovaFabric install, which is what makes them evidence.
The standards stack is named and checkable (OTel GenAI, OpenLineage, in-toto DSSE, RFC 3161), with a conformance page listing how to check each claim.

Cautions
#

7 stars, 1 fork, and no independent coverage: the verification story is the project’s own conformance claims.
Capsule and bundle formats are additive but not frozen until v1.0, so archived capsules may need migration.
The secret scan matches known key formats only, so a capsule is scanned, not clean, by the README’s own wording.
Server mode, dashboard, and cluster collection are experimental; the single-player path is the usable one.
One maintainer with a consultant’s day job, at v0.108 in five months, is a pace that can stop as fast as it started.

Pricing
#

Free and open source under Apache-2.0; no paid tier, no hosted service, and the core works fully offline.

Compared to
#

Agent-Causality: the causal sibling; Agent-Causality slices recorded runs for cause, NovaFabric preserves and diffs whole runs for proof.
claude-replay: the other replay in this category; claude-replay renders a session for humans to watch, NovaFabric replays calls for machines to verify.
agentsview: the archive that reads the session files tools already wrote; NovaFabric needs to be in the run’s path (loader or proxy) but captures what those files never hold.

Bottom line
#

Recommended for engineers who must reproduce or prove what a run did, weeks later and offline, and for gating agent behavior change in CI. Not for live cost dashboards, transcript search, or anyone who needs a frozen evidence format today.

Changes
#

2026-10-28 - Created from the coordinator’s entrant handoff, admitted on the Run Capsule mechanism, the five replay modes with the diff CI gate, and eight fetched sources.

References
#

https://api.github.com/repos/MSKazemi/novafabric - 7 stars, 1 fork, Apache-2.0, created 2026-06-25, pushed 2026-10-26, as of 2026-10-28
https://raw.githubusercontent.com/MSKazemi/novafabric/main/README.md - the capsule layout, capture paths, five replay modes, diff gate, sealing, comparison table, and status labels
https://api.github.com/repos/MSKazemi/novafabric/releases - v0.108.0 (2026-10-25), v0.107.0 (2026-10-23), v0.106.1 (2026-10-23)
https://pypi.org/pypi/novafabric/json - PyPI package, version 0.108.0, Python 3.12 or newer
https://pypistats.org/api/packages/novafabric/recent - 734 downloads in the trailing month to 2026-10-28, behind the adoption claim
https://novafabric.ai/docs/ - the documentation hub this run verified live
https://arxiv.org/abs/2609.12582 - the paper on tamper-evident, replayable evidence for agent runs
https://hn.algolia.com/api/v1/search?query=novafabric&tags=story - zero stories, the empty independent footprint this note states

run-recording on tomrochette.com

Agent-Causality

What it is
#

Status
#

Strengths
#

Cautions
#

Pricing
#

Compared to
#

Bottom line
#

Changes
#

See also
#

References
#

NovaFabric

What it is
#

Status
#

Strengths
#

Cautions
#

Pricing
#

Compared to
#

Bottom line
#

Changes
#

See also
#

References
#

run-recording on tomrochette.com

Agent-Causality

What it is #

Status #

Strengths #

Cautions #

Pricing #

Compared to #

Bottom line #

Changes #

See also #

References #

NovaFabric

What it is #

Status #

Strengths #

Cautions #

Pricing #

Compared to #

Bottom line #

Changes #

See also #

References #

What it is
#

Status
#

Strengths
#

Cautions
#

Pricing
#

Compared to
#

Bottom line
#

Changes
#

See also
#

References
#

What it is
#

Status
#

Strengths
#

Cautions
#

Pricing
#

Compared to
#

Bottom line
#

Changes
#

See also
#

References
#