Show HN：Ctrl – AI 代理的執行控制平面

Navigation Menu

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

To see all available qualifiers, see our documentation.

Execution control plane for AI agents.

Uh oh!

There was an error while loading. Please reload this page.

MehulG/agent-ctrl

Folders and files

Latest commit

History

Repository files navigation

ctrl

Execution control plane for AI agents.

Ctrl sits between agent intent and real-world actions.
Agents can decide what to do. Ctrl decides what’s allowed to happen.

It intercepts tool calls, risk-scores them, enforces policy (allow / deny / approve),
and executes only what’s authorized — with a full, auditable ledger.

One-liner: a drop-in CtrlMCP wrapper for LangChain that turns agent actions into
governed execution.

Status: early v0; APIs may change quickly — treat demos as the source of truth.

Why Ctrl exists

AI agents are moving from reading and drafting to acting:
sending emails, issuing refunds, publishing content, changing production systems.

The moment agents take real actions, intelligence stops being the bottleneck —
authority, safety, and auditability do.

Today, teams solve this with ad-hoc allowlists, brittle checks, and manual approvals.
That doesn’t scale, and it breaks the moment agents run faster than humans.

Ctrl is the missing layer:
a runtime that decides whether an agent action should happen,
under what constraints, and with what proof.

Think of Ctrl as an action gateway:
agents propose actions, Ctrl authorizes and executes them safely.

Demo

5-minute demo (publish a market report)

Run an end-to-end demo where an agent fetches crypto data and attempts to publish
a static page via EdgeOne.

The publish action is intercepted, risk-scored, paused for approval, and
replayed safely after approval.

If policy returns pending, the agent currently exits after printing a request ID.
This is intentional: nothing runs until approval is recorded.

Start the approvals API (same configs + shared ctrl.db):

Start the dashboard UI locally:

Approvals API: http://localhost:8788
Dashboard UI: http://localhost:3000

Docker (agent + approvals + dashboard)

Prefer containers? Use the bundled compose in
demos/e2e_publish_market_report/.

Approvals API: http://localhost:8788
Dashboard UI: http://localhost:3000

What Ctrl does today

What Ctrl is evolving into

A general-purpose action gateway for AI agents.

The long-term goal is not approvals UI —
it’s making autonomous execution safe by default:

As agents begin handling customer communications and financial actions,
Ctrl becomes the place where organizations decide:

Which actions can run automatically — and which must never run unchecked.

Architecture

Requirements

Install

Quickstart: LangChain + MCP

1) Define servers and policies

configs/servers.yaml

configs/policy.yaml

configs/risk.yaml

2) Initialize the database

3) Wrap your MCP client

Tool calls now flow through:
log intent → score risk → decide policy → enforce → execute.

More example for policy/risk/server bundles live in docs/example-policies/.

Audit trail (SQLite)

Example:

Development

PRs and issues are welcome. Keep changes small and reliability-focused.

About

Execution control plane for AI agents.

Resources

Uh oh!

There was an error while loading. Please reload this page.

Stars

Watchers

Forks

Releases

Packages

  0

Languages

Footer

Footer navigation