oss-b-011record 1
{
"rank": 1,
"id": "oss-b-011",
"identity": "ActivityWatch/activitywatch",
"url": "https://github.com/ActivityWatch/activitywatch",
"segment": "task-mining",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Pragmatic tier-1 sensor: app/tab telemetry answers the questions that burn the first client meeting, at low consent cost",
"mechanism": "The lowest-invasiveness, highest-maturity capture option: app names, window titles and tab URLs are enough to infer which systems a client actually lives in and roughly how their day splits, without recording screen content or keystrokes. This is the pragmatic tier-1 sensor for precomputing client context.",
"limitations": "Gives app-level attention data, not process structure — you learn they spend four hours in HubSpot, not what workflow they run inside it. Window titles leak more than users expect (document names, customer names, subject lines); one adapted fork exists specifically to disable window-title tracking, which is itself evidence that the default is too revealing. MPL-2.0 is file-level copyleft: modifications to ActivityWatch files must be published, though it will not infect surrounding proprietary code.",
"license_claimed": "MPL-2.0",
"disposition": "top10_promoted"
}
oss-b-044record 2
{
"rank": 2,
"id": "oss-b-044",
"identity": "getzep/graphiti",
"url": "https://github.com/getzep/graphiti",
"segment": "org-graph",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Bi-temporal invalidation plus episode provenance — the right substrate for a client-context store",
"mechanism": "The best-fitting org-graph primitive found, for two reasons specific to client intelligence. Bi-temporal invalidation means when a client's situation changes the old fact is superseded rather than deleted, so the system knows what it used to believe and when that stopped being true. Episode provenance means every precomputed claim traces to the source document or conversation that produced it — the same defensibility property that makes fire-enrich's per-field citations valuable, but built into the data model.",
"limitations": "Open-core: it is the engine under Zep's paid product, so expect the roadmap to serve that. Needs a graph database — Neo4j 5.26, FalkorDB, or Amazon Neptune plus OpenSearch Serverless; the lightweight embedded option (Kuzu) is explicitly deprecated because upstream is unmaintained, so there is no cheap local backend. Requires an LLM API key per ingestion, so graph construction has a running per-client cost. A 206-open-PR backlog against 285 open issues signals a project moving faster than it is being reviewed.",
"license_claimed": "Apache-2.0",
"disposition": "top10_promoted"
}
oss-b-036record 3
{
"rank": 3,
"id": "oss-b-036",
"identity": "scikit-activeml/scikit-activeml",
"url": "https://github.com/scikit-activeml/scikit-activeml",
"segment": "active-learning",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Stream-based active learning with a labelling budget is the correct frame for question selection under client patience",
"mechanism": "The better-maintained and better-licensed sibling of modAL for question-selection policy. Its stream-based strategies matter more than the pool-based ones for Actionist: a client conversation is a stream where each potential question arrives and must be asked-or-skipped immediately under a budget, which is exactly the stream-based active-learning formulation with a labelling budget.",
"limitations": "Its own README describes it as a library for research; expect research ergonomics rather than a production policy engine. Strategy count varies by source (search said 47 pool + 14 stream, the repo says 60+) — quote the repo's 60+ or, better, count the strategy overview page before quoting anything. 201 stars is a small community for a library carrying this much surface area. Same conceptual gap as modAL: uncertainty is a proxy for question value, not a measure of it.",
"license_claimed": "BSD-3-Clause",
"disposition": "top10_promoted"
}
oss-b-024record 4
{
"rank": 4,
"id": "oss-b-024",
"identity": "grokify/traffic2openapi",
"url": "https://github.com/grokify/traffic2openapi",
"segment": "introspection",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Authority arbitrage: a HAR export the client makes in 30 seconds yields API evidence needing no installed agent",
"mechanism": "The low-authority alternative to APIClarity: a HAR file exported from the client's own browser DevTools during a normal working session reveals which internal and third-party APIs their tools call, with no agent installed and no cluster access. That is an unusually cheap way to precompute a client's real integration map.",
"limitations": "3 stars and zero forks — a single-maintainer, pre-release project with essentially no production validation; the engineering hygiene is genuine but is not adoption. HAR captures contain live auth tokens, cookies and response bodies, so the artifact itself is a credential-bearing secret that must be scrubbed before it is stored or shared. Inference from a single session sees only the endpoints that session happened to touch.",
"license_claimed": "MIT",
"disposition": "top10_promoted"
}
oss-b-009record 5
{
"rank": 5,
"id": "oss-b-009",
"identity": "OpenAdaptAI/OpenAdapt",
"url": "https://github.com/OpenAdaptAI/OpenAdapt",
"segment": "task-mining",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Halt-don't-interpolate discipline; deterministic workflow bundle rather than a pixel macro",
"mechanism": "The strongest task-mining candidate. Its explicit design claim — zero generative-model calls on healthy runs, halt instead of guess — is the discipline Actionist wants when inferring a client's process from observation: produce an inspectable artifact, refuse to hallucinate the gaps.",
"limitations": "Mid-refactor, so the API surface is unstable and the legacy monolith is frozen rather than maintained; adopting now means tracking a moving multi-repo target. Screen+keyboard+OCR capture on a client machine is the maximum-invasiveness end of the spectrum. The PII/PHI scrubbing lives in a separate optional extra, meaning the safe configuration is opt-in, not default.",
"license_claimed": "MIT",
"disposition": "top10_promoted"
}
oss-b-010record 6
{
"rank": 6,
"id": "oss-b-010",
"identity": "xlang-ai/OpenCUA",
"url": "https://github.com/xlang-ai/OpenCUA",
"segment": "task-mining",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Action reduction plus review-trim-submit consent gate — consent to the artifact, not to the process",
"mechanism": "AgentNetTool is the best-licensed demonstration recorder found: synchronized screen video, mouse/keyboard events AND accessibility trees, plus a browser-based review/trim/submit step. That human review gate before submission is the exact consent pattern Actionist should copy — the client sees and edits what gets captured before it leaves the machine.",
"limitations": "The recorder is instrumental to a model-training agenda, not a product; expect research-grade ergonomics and thin support. Accessibility-tree capture pulls in whatever text is on screen, including material the client never meant to share. Dataset claim varies by source (22.5K vs 22.6K) — quote the repo's 22.6K.",
"license_claimed": "MIT — README explicitly states use permitted for research, educational AND commercial purposes",
"disposition": "top10_promoted"
}
oss-b-001record 7
{
"rank": 7,
"id": "oss-b-001",
"identity": "process-intelligence-solutions/pm4py",
"url": "https://github.com/process-intelligence-solutions/pm4py",
"segment": "process-mining",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Conformance checking as the precompute asymmetry (described vs actual process); AGPL + commercial upsell",
"mechanism": "The canonical way to turn a client's raw system logs into an observed process model without asking them to describe their process. If Actionist can obtain any timestamped event stream (CRM stage changes, ticket transitions, order states), pm4py yields the as-is workflow, rework loops and bottlenecks as evidence rather than self-report.",
"limitations": "AGPL-3.0 is the trap: linking pm4py into a hosted Actionist service arguably triggers the network-copyleft clause, and the vendor sells a commercial license precisely because of this. Also requires a case ID + activity + timestamp triple that most SMB clients do not have lying around; log extraction is the real work, not the mining.",
"license_claimed": "AGPL-3.0 — dual-licensed; README explicitly offers a separate paid license for commercial use in closed-source environments",
"disposition": "top10_promoted"
}
oss-b-016record 8
{
"rank": 8,
"id": "oss-b-016",
"identity": "firecrawl/fire-enrich",
"url": "https://github.com/firecrawl/fire-enrich",
"segment": "enrichment",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Per-field source citation, the most portable idea in the survey; MIT shell over rented paid capability",
"mechanism": "The single most directly transferable pattern in this survey: given nothing but a client's email address, produce a cited company dossier before the first conversation. The per-field source-URL citation is the design detail that makes precomputed context defensible — Cena can see where each claim came from instead of trusting a model.",
"limitations": "MIT covers the orchestration shell only; the actual capability is rented — it will not run without a Firecrawl API key and an OpenAI key, so both cost and data flow sit with third parties. 35 commits is a demo-scale codebase, essentially a showcase for Firecrawl's paid API (compare the Composio pattern: permissive licence over a paid hosted dependency). Synthesis is GPT-4o-mediated, so field values are model outputs, not scraped facts.",
"license_claimed": "MIT",
"disposition": "top10_promoted"
}
oss-b-004record 9
{
"rank": 9,
"id": "oss-b-004",
"identity": "bupaverse/pm4py",
"url": "https://github.com/bupaverse/pm4py",
"segment": "process-mining",
"observed": "2026-08-27",
"evidence_class": "I",
"verified": "search-only",
"denominator": 50,
"selection_reason": "Permissive escape hatch from pm4py's AGPL — licence UNCONFIRMED, must be read from DESCRIPTION",
"mechanism": "Only interesting as evidence that the two ecosystems interoperate; would let an R-based pipeline reach pm4py's conformance algorithms.",
"limitations": "Not fetched — stars, licence, and activity all unverified. Critically, wrapping PM4Py does not escape PM4Py's AGPL-3.0; the bridge inherits the upstream licensing question rather than solving it.",
"license_claimed": "not observed",
"disposition": "top10_promoted"
}
oss-b-023record 10
{
"rank": 10,
"id": "oss-b-023",
"identity": "openclarity/apiclarity",
"url": "https://github.com/openclarity/apiclarity",
"segment": "introspection",
"observed": "2026-08-27",
"evidence_class": "D",
"verified": "fetched",
"denominator": 50,
"selection_reason": "Best-in-segment introspection tool retained as a negative finding: archived 29 May 2026, now unmaintained CVE surface",
"mechanism": "The clearest statement of the introspection pattern in the whole survey: infer the true integration surface from observed traffic, then treat the gap between documented and observed as the finding. That gap — what the client's systems actually do versus what anyone documented — is exactly the value Actionist claims to deliver.",
"limitations": "Archived three months ago, so unmaintained and accumulating unpatched CVEs in a security-adjacent tool. Assumes a Kubernetes/service-mesh estate (Istio 1.10+, DaemonSet tap, Kong, Tyk or OTel collector); can run with ENABLE_K8S=false but that is the degraded path. Requires deploying a traffic tap inside the client's production cluster — a very high authority and trust bar, and it observes real request/response payloads including customer data.",
"license_claimed": "Apache-2.0",
"disposition": "top10_promoted"
}