Make your agents show their work. Kontour makes AI-assisted work inspectable and accountable: sources stay traceable, claims keep their evidence, required paths advance through explicit gates, and people can recompute why work is ready.
→ kontourai.io tells the product story, and kontourai.github.io indexes current documentation for its listed products. The repositories below are the public implementation and documentation surfaces.
| Repository | Responsibility |
|---|---|
station |
Open-source, local-first agent workspace: projects, agents, and devices in one place, with gates, evidence, and receipts beside the work (station.kontourai.io) |
fieldwork |
Turns documents into reviewed, source-grounded evidence you can recheck and export as a portable trust bundle |
| Repository | Responsibility |
|---|---|
flow-agents |
Portable workflow engine, runtime adapters, hooks, and first-party kits for evidence-gated agent work |
veritas |
Executable repository standards and evidence-backed merge readiness for AI-authored code |
surface |
Kontour's integration surface for portable claims, evidence, trust status, validation, and product-facing Hachure compatibility |
flow |
Process definitions, steps, gates, evidence expectations, transitions, and explicit exceptions |
survey |
Producer-side source, extraction, candidate, review, decision, and publication records |
console |
Suite operating plane: claim, process, proof, and queue state across the products, without owning their semantics |
The Builder Kit and Knowledge Kit are first-party solutions distributed
through Flow Agents rather than separate products.
aidlc-kit runs the pinned AWS AI-DLC
methodology as an independently maintained Flow Agents kit. The open trust format
itself lives independently at hachure-org/spec;
Surface is Kontour's product-facing integration layer for it.
| Repository | Responsibility |
|---|---|
forage |
SSRF-pinned web acquisition, provenance-bearing snapshots, and deterministic replay |
traverse |
Schema-directed, provenance-bearing extraction proposals |
lookout |
Registered-source rechecks, source drift, and deterministic proposal diffs |
bearing |
Evidence-backed model capability observations, deterministic catalogs, and runtime-aware ranking |
datum |
Provider, model, secret-reference, and role configuration resolution |
relay |
Provider-neutral model invocation contracts, adapters, and conformance tools |
dispatch |
Policy-aware model execution with routing, budgets, fallbacks, and receipts |
conduit |
Portable agent-host contracts: lifecycle projection, asset installation, and conformance evidence |
plumb |
Deployment health checks that hand failures to an agent in an isolated clone for a reviewable PR or issue |
ephemeris |
Time actor: deadline wake-ups that nudge Flow, plus a timezone-aware recurring-schedule core |
ui |
Shared design tokens, React primitives, and web components for Kontour product interfaces |
cli |
The offline-first kontour command that routes to installed Kontour products |
thread |
Portable conversation schema and Ferry transcript conversion, with adapter-specific fidelity limits (documentation) |
corpus |
Source-bound documentation checking, GitHub Pages publishing, and docs MCP retrieval |
kit-research is a third-party kit
authoring experiment. Its README states its current lifecycle and limits.
- File issues in the repository that owns the behavior or contract. Start at kontourai.io/developers if you are unsure which layer owns a concern.
- Read the target repository's
AGENTS.md,CONTEXT.md, contribution guide, and architecture decisions before changing product language or contracts. - Check each repository for its license, supported runtime, release status, and verification commands; those details intentionally live with the owning code.
We are looking for teams that want to:
- run Station or Flow Agents on real development work and measure whether the required path improves delivery quality, cost, and continuity;
- use Veritas to turn repository standards into evidence an agent and reviewer can both inspect; or
- build a product integration through Surface while keeping the underlying Hachure records portable and independently recomputable.
Email hello@kontourai.io with the workflow you want to make inspectable.
Evidence-backed confidence. Not certainty theater.