acp-ai 0.2.0

Agent Control Plane — SDK and CLI for governed agent execution with a local Docker stack.

ACP Control Plane — Python SDK + CLI

PyPI: acp-ai · Import: acp

One pip install gives you a local governance stack: intercept tool calls, evaluate your OPA/Rego rules, persist audit + agent registry in Postgres, and view everything in the dashboard.

Requires: Docker Desktop (Compose v2). All acp commands run in your terminal (Terminal.app, iTerm, or the IDE integrated terminal).

---

Install (terminal)

pip install acp-ai
acp doctor

---

First-time setup (terminal)

Run these in order from any directory (home folder is fine):

# 1. Create your editable policy folder (~/.acp/policies)
acp init

# 2. Start local stack: Postgres + OPA + interceptor + dashboard
acp up --build

# 3. Open governance dashboard in browser
acp dashboard

Command	What it does	Where it runs
acp init	Creates ~/.acp/policies/ with starter .rego files + HOW_TO_ADD_RULES.md	Your terminal
acp up --build	Starts Docker containers (first time builds interceptor image)	Your terminal (needs Docker running)
acp up	Starts stack without rebuild (after first --build)	Your terminal
acp down	Stops containers	Your terminal
acp status	Container + health check status	Your terminal
acp dashboard	Opens http://localhost:3090/dashboard/	Your terminal
acp doctor	Checks Docker + package install	Your terminal

---

Your rules (editable after pip install)

Policies are yours, not locked inside the package:

~/.acp/policies/
├── HOW_TO_ADD_RULES.md
├── router.rego                 ← route tools → policies
├── rbac.rego                   ← JWT roles → tools
├── mortgage_application.rego   ← commented sample (loan submit)
└── mortgage_underwriting.rego  ← commented sample (score_loan / escalate)

Edit files in ~/.acp/policies/, then restart:

acp down && acp up

Custom policy path (optional):

export ACP_POLICIES_DIR=/path/to/my/policies
acp up

---

Use in your agent project (any framework)

In your project folder (separate from ACP), activate a venv and use the SDK:

import os
from acp import governed_tool

os.environ.setdefault("ACP_INTERCEPTOR_URL", "http://localhost:8080")

@governed_tool(agent_id="my-agent", tool="my_tool_api", action="run_action")
def my_tool(amount: int) -> str:
    return f"ok:{amount}"

Works with CrewAI, Strands, LangGraph, or plain Python.

Register agents via dashboard Agent Registry or POST /api/v1/agents.

---

Local URLs

URL	Purpose
http://localhost:3090/dashboard/	Dashboard (events, policies, registry, approvals)
http://localhost:8080	Interceptor API (/tool-call, /api/v1/*)

---

What Docker starts

Service	Role
Postgres	Audit log, agent registry, approvals
OPA	Evaluates your Rego from ~/.acp/policies/
Interceptor	PEP — POST /tool-call, JWT, registry
Dashboard	Governance UI
Gateway	Single origin on :3090

Also on disk: ~/.acp/jwt/ (dev RS256 keys), ~/.acp/runtime/ (compose + images).

---

Maintainers (ACP monorepo)

cd sdk/python
./scripts/sync_bundle.sh
./scripts/publish_pypi.sh
twine upload dist/*