v0.1.10 npm install -D @kontourai/flow

Flow

required paths, gates, evidence, and exceptions made inspectable

Agents skip steps, accept weak evidence, summarize work as complete, and lose the thread after compaction. Flow is the small thing missing in the middle: a record of the required path, the evidence each gate expected, the evidence that was actually collected, and the exceptions that need explicit human trust. It does not replace workflow engines, agent runtimes, observability tools, approval apps, or policy engines. It explains why the work was allowed to advance.

View on GitHub → Built with Surface
flow report — current run

$ flow status run.dev-1847 --format summary


flow run: agent-dev-flow / feature-search-filters

current step: verify


PASS plan gate: acceptance criteria linked

PASS implementation gate: scoped files changed

BLOCK verify gate: browser evidence missing

expected: tests, lint, screenshot, Veritas readiness


next action: run browser check before publish

continuation: resume from verify, not chat memory

report: .flow/runs/run.dev-1847/report.md

Quickstart

flow quickstart

$ npm install -D @kontourai/flow

$ npx flow init

$ npx flow start examples/agent-dev-flow.json --run-id dev-1847

$ npx flow attach-evidence dev-1847 --gate verify-gate \

--file ./test-output.json --kind command

$ npx flow report dev-1847

What Flow answers

01 What process path was required?
02 Which step is active now?
03 Which gate controls the next transition?
04 What evidence does the gate expect?
05 What evidence was actually collected?
06 Why was the transition allowed or blocked?
07 Was an exception accepted, and by whom?
08 How should another agent resume?

Example use case

A release path that waits for evidence.

A team asks an agent to ship a pricing-page change. Flow records the required path: plan, implementation, code review, rendered-page verification, release readiness, and post-deploy smoke. The work can only advance when each gate has the evidence it asked for.

If the browser check is missing, the run does not become "done." It stays blocked with a resumable next action.

release-readiness-flow

PASS plan: acceptance criteria present

PASS implementation: scoped diff

PASS review: no high findings

BLOCK verify: rendered-page screenshot missing

expected: desktop + mobile evidence


next action: capture browser evidence, then resume

The difference

A trace says what happened.
Flow says why it was enough.

Observability is necessary, but a trace alone does not know whether the right verification happened, whether approval had enough context, or whether the agent skipped a required handoff.

Flow turns required paths into evidence-gated transitions. If the evidence is missing, the run blocks or routes back. If the user accepts the gap, the exception becomes part of the record.

Flow Definition
The required path and gate expectations
Flow Run
One concrete execution with state and history
Gate Evidence
Tests, CI, traces, approvals, Veritas reports
Flow Report
Current state, blocked gates, next action

What Flow does not replace

Workflow engines

Camunda, Temporal, Dapr, Inngest, and Trigger.dev can still run durable work. Flow records gate state and evidence.

Agent frameworks

LangGraph, CrewAI, OpenAI Agents, Microsoft Agent Framework, and runtime SDKs can still execute agent loops.

Observability

Phoenix, LangSmith, AgentOps, and OpenTelemetry can still store traces. Flow interprets traces against gate expectations.

Policy systems

OPA, security tools, and Veritas can still evaluate rules. Flow records their decisions as process evidence.

Flow Console

See the run
where the work happens.

flow console --run <id> serves a live view of a run straight from .flow/runs/<id>/. Loopback only. No cloud, no login.

Process graph
See the required path and where the run sits on it right now.
Transition timeline
Walk every gate the run cleared, blocked on, or routed back from.
Gate detail
Inspect the evidence a gate expected and what was actually collected.
Evidence links
Jump from a gate to the tests, CI, traces, or readiness reports behind it.
flow console

$ npx flow console --run dev-1847


Flow Console running at http://127.0.0.1:4317


Reading: .flow/runs/dev-1847/

run: agent-dev-flow / feature-search-filters


GET / 200 console shell

GET /api/run 200 run graph + timeline

GET /api/gates 200 gate detail + evidence


───────────────────────────────

PASS plan · implementation

BLOCK verify · browser evidence missing

loopback only · no hosted service

Product line

One transparency stack.
Independent products.

Surface makes claims inspectable. Survey makes producer-side provenance inspectable. Flow makes process transitions inspectable. Flow Agents applies Flow and Veritas discipline inside Codex, Claude Code, Kiro, GitHub Actions, and future agent tools.

You get a workflow that simply works. Underneath, each Kontour product adds one focused layer of the trust stack.

kontour stack

Surface claims, evidence, policies, trust snapshots

Survey producer-side source → review → claim, projected into Surface

Flow steps, gates, transitions, exceptions, reports

Veritas repo standards, evidence checks, merge readiness

Flow Agents runtime-native workflows + work modes


shared promise: Kontour AI shows the work behind AI.