Open source · MIT · local-first · works with Claude Code & Cowork

Put your AI to work —
and keep it accountable.

Sentigent gives Claude Code and Claude Cowork what they're missing: a company to serve, a plan to finish, and a memory that learns from real results — so your team's AI ships verified work instead of guesses.

$pip install sentigent
add it to the agent you already use
# 1 · plug it into Claude Code (or any AI coding agent)
pip install sentigent
sentigent init          # MCP + hooks, one time

# 2 · keep coding like normal. now your agent…
 remembers the plan across sessions
 keeps going at a blocker instead of stopping to ask
 still stops for the dangerous calls  # force-push · prod · secrets

# when a run ends — one honest number, not a vibe
FAP 0.88  # how far it got with no help
The north star · where this goes

AI frontier teams — for your company.

The harness frontier AI labs build for their own agents — company knowledge, an accountable plan, a self that learns — given to any small team, on top of Claude Code & Cowork. They already act in your real tools (Gmail · Drive · Slack · Stripe…) over MCP; Sentigent supplies the three things that turn a tool into a colleague, then learns from every outcome. One MCP server, useful from every Claude surface.

① ORGANIZATION

A company to serve

North-star, mission, brand, policy, playbooks as one shared, drift-free knowledge bundle (OKF). The worker stays on-mission and in-policy.

status: building
② PROJECT

A plan to follow

A goal becomes a phased, verified plan the loop drives to done across sessions — reported as FAP, one honest number.

status: live
③ AGENT

A self to work as

A shadow agent per teammate that learns their judgment and earns autonomy from outcomes — shadow → suggest → draft → own the loop.

status: core live · ladder building
# the whole picture — one loop
 you set direction  ──▶  SENTIGENT supplies  ORGANIZATION · PROJECT · AGENT  ──▶  CLAUDE CODE / COWORK acts
                                                                                   (plans · edits · sends · files, via MCP)
                              ▲                                                              │
                              └────────────────  outcome graded  ◀──────────────────────────┘
                                  record_outcome → the brain learns → autonomy earned, drift caught
 legend:  ✅ verified, no help    ○ escalated to you    ❌ blocked by policy   (rolled up as FAP)

The execution layer is commoditizing — that's good for us. What compounds with use is the layer we own: a company's knowledge, a verified plan, an employee that learns. Full direction → PRODUCT-DIRECTION.md

Honest about enforcement: on Claude Code, Sentigent can block a risky action before it runs (policy, not a model's guess). On Claude Cowork, it brings your company's knowledge and records every outcome today — policy-level action-blocking is rolling out as we extend enforcement across surfaces.

Why it finishes

Build your way. Deliver outcomes that work. Stay in control.

Three things a raw loop can't do on its own — and the reason Sentigent doesn't need babysitting and won't drive off a cliff.

01 — INDIVIDUAL · PROFILE

Build your way

Sentigent learns your push-vs-ask judgment from the blockers you answer, writes it back as a steering file your agent reads — and enforces the practices you pick (tests before a commit, review before a merge) at the moment they matter. The more it sees, the less it asks.

sentigent practices → enforced, not just advised
# pick the practices to enforce — off · warn · block
sentigent practices add "Run the tests before committing"
sentigent practices enforce 1 block

# commit with no test run this session →
git commit -m "wip"
⛔ blocked  # practice not met · run tests, or drop it to warn
02 — PROJECT · PLAN

Deliver outcomes that work

Every step carries its own done-criteria, and nothing counts until it verifies. When the run ends you get FAP — Faithful Autonomous Progress — one honest number, not a made-up confidence score.

per-step gates → FAP receipt
# each step has its own gate; fail → self-repair retry
steps:
  - write parser   verify: pytest tests/test_parse.py
  - add tests      verify: pytest -q
  - wire the CLI   verify: sentigent --help

sentigent loop receipt
FAP = verified-with-no-help ÷ total  # real, per run
03 — ORG · POLICY

Stay in control

A hard wall the loop can never auto-clear. Force-push, production databases, secrets, rm -rf, anything that leaves your machine — those always stop and ask you. Humans set intent; the agent acts; you gate the rest.

guardrails/*.yaml
# never auto-cleared — always escalates to a human
escalate:
  - force_push
  - prod_db_write
  - secret_access
  - rm_rf
  - external_send

# the loop pauses, asks you, and remembers your answer.
One run, step by step

The loop that survives the end of a session

State lives on disk in ~/.sentigent/loops/. Kill the process, hit a usage limit, close the laptop — a fresh agent reloads the plan and its position and keeps going.

01

Plan

Goal + steps + per-step done-criteria, stored durably.

02

Act

A fresh agent runs the next step with full context.

03

Verify

Re-runs your done-criteria. Fail → self-repair retry.

04

Decide

Go or ask — as you would. Dangerous calls always ask.

05

Resume

Reloads plan + position from disk. Survives restarts.

What people use it for

Long jobs that outlast a single context window

Drive a plan to done

Hand it a vision and a plan; it works lap after lap until the done-criteria are met.

Survive session limits

The plan and its position are on disk, so a context limit or crash is just a pause.

Learn from every blocker

Each question you answer becomes a precedent — so next time it answers it itself.

The model

Three layers of context decide every call

Each layer owns a different question. Together they answer the only one that matters at a blocker: keep going, or stop and ask?

Yours · free, local

Individual · Profile

  • A model of how you work
  • Your push-vs-ask judgment
  • Enforces the practices you pick
This project · free, local

Project · Plan

  • Goal + plan + done-criteria
  • Repo guardrails
  • Durable across sessions
Your team · free, self-host

Org · Policy

  • Hard rules that never bend
  • Approved tools & budgets
  • One audit trail
decide = Profile · Plan · Policy

See the full picture in diagrams →

Pricing

Free and open source — top to bottom.

All three layers run on your machine for free, forever. If a team would rather not self-host the org control plane, a managed version will come later — but everything works free today.

Open source
$0 / forever

The whole thing — nothing held back.

  • The full loop — Plan → Act → Verify → Decide → Resume
  • Individual · Profile (your judgment, your steering file)
  • Project · Plan with per-step done-criteria
  • Org · Policy — rules, registry, budgets, audit trail (self-host)
  • The policy wall on dangerous actions
  • FAP receipts · local SQLite memory · MIT licensed
Install it
Managed
Later · optional, not yet

For teams who'd rather not run the org control plane themselves.

  • Hosted Org · Policy control plane
  • Shared org knowledge base across projects
  • Parallel / async runs from a backlog
  • Everything here is already free to self-host
  • We'll only launch this if teams ask for it
Follow along
Quickstart

Two commands to wire it into your agent

your terminal
# 1 — install
pip install sentigent

# 2 — wire it into Claude Code / any AI coding agent
sentigent init

# now seed a loop and let it run
sentigent loop start --goal "…" --steps "…" --verify "pytest -q"
sentigent loop drive <loop_id> --execute