# Meetless (mla) > An open-source CLI that governs AI coding agents by capturing decisions, detecting stale or conflicting instructions, and injecting human-approved context before each agent action. Meetless (mla) is an open-source command-line tool built by An Pham that acts as the active source of truth for AI coding agents like Claude Code and OpenAI Codex. It sits between the developer and their coding agents, watching sessions, capturing decisions as work happens, detecting conflicts, and proactively injecting human-approved context before each agent turn. The project is currently in public beta, free to use, and licensed under Apache-2.0. ## What It Is `mla` (Meetless Agent) is a governance control layer for AI coding agents. Rather than relying on static files like `CLAUDE.md` or retrieval-augmented memory that cannot distinguish which of two conflicting decisions is still in force, `mla` maintains a structured record of human-approved decisions and injects them into every agent session automatically. The core insight the project is built around: the problem with coding agents is not missing memory, it is missing *authority* — the ability to know which decision is currently in force. ## How the Governance Loop Works `mla` runs a four-step loop on every governed session: - **Watch** — installs as a session hook so capture is a side effect of working, not a separate chore - **Capture** — records decisions with evidence as they happen, not buried in transcripts - **Detect** — surfaces conflicts when a new session contradicts an already-settled decision, or when a file in the repo still asserts something a ruling replaced - **Steer** — before the agent acts, injects the decisions still in force into its context via a structured `` block Nothing an agent proposes becomes project truth until a human approves it. Unresolved conflicts are routed to a human inbox rather than left for the agent to guess at. ## Architecture and Deployment Model The `mla` client is a TypeScript CLI distributed as a self-contained binary (no Node required for the one-liner or Homebrew installs; Node 22+ required for the npm package). It connects to two backends: `control` (the system of record) and `intel` (the AI runtime). The prompt-time injection path is a local read of local state and requires no network request. Session capture — the prompts, decisions, tool calls, and documents from governed sessions — is sent to the backend for the governance service to function. Supported agents and wiring: - **Claude Code** — `UserPromptSubmit` floor injection, MCP retrieval via `meetless-mcp`, `PreToolUse` enforcement - **OpenAI Codex** — same neutral decision core, same injection and enforcement model, separate connector registration The MCP server (`meetless-mcp`) also allows any MCP-capable agent to read governed memory directly. Platforms supported are macOS (Apple Silicon, prebuilt binary) and Linux (x86_64 glibc); Windows is community-supported via WSL. ## Research and Benchmarking Approach The Meetless site publishes a suite of internal benchmarks comparing governed memory against static `CLAUDE.md` files and RAG-based retrieval. The benchmarks test freshness (whether the agent identifies the decision currently in force after a supersession), provenance (whether the agent can report who approved a decision and when), and conflict detection (whether contradictory decisions from separate sessions are caught before they land). The project states that three results have been withdrawn and one narrowed, all on its own initiative, and that one benchmark built to prove the core thesis disproved it — that result is published as well. The site notes these are preliminary internal benchmarks at small N and labels them accordingly. ## Update: mla 0.2.27 The latest release is `cli-v0.2.27` (mla 0.2.27), published on 2026-07-24. The repository was created in June 2026 and last pushed in August 2026, indicating active development. The Codex connector — including the `UserPromptSubmit` wrapper, static plugin package, `mla codex install`/`uninstall` commands, and `mla doctor` health checks — was added as a recent major surface, built using Codex CLI 0.144.6 running GPT-5.6. The project is currently in beta with no paid plans set; the pricing page states plainly that pricing has not been determined yet and that paid plans will be announced when they open. ## Why It Matters for Agent-Driven Development As coding agents take on longer-running tasks across multiple sessions, the gap between what a team has decided and what an agent acts on widens. `mla` addresses this by treating architectural decisions as first-class governed artifacts — analogous to how git is the system of record for code — rather than leaving them in files the agent may or may not read, or in memory systems that cannot resolve which of two retrieved answers is still authoritative. The tool is designed for both solo builders and software teams, with team workspaces allowing shared owner-approved decisions to be delivered to every connected agent session. ## Features - Session hook-based decision capture — no manual chore - Human approval gate before any agent proposal becomes project truth - Conflict detection across sessions and stale files - Per-turn context injection via structured blocks - MCP server for governed memory retrieval by any MCP-capable agent - Claude Code and OpenAI Codex connectors sharing one neutral decision core - Conflict inbox routing unresolved authority disputes to humans - Enforcement ceiling (observe/warn/ask/deny) configurable per session - Team workspaces with shared owner-approved knowledge - Open-source client (Apache-2.0) with inspectable telemetry - mla doctor health checks for connector and auth status - Ed25519-signed update manifests - MEETLESS_TELEMETRY=off opt-out for all telemetry planes ## Integrations Claude Code, OpenAI Codex, MCP (Model Context Protocol), npm, Homebrew, WSL (Windows Subsystem for Linux) ## Platforms WINDOWS, MACOS, LINUX, WEB, API, CLI ## Pricing Open Source ## Version 0.2.27 ## Links - Website: https://meetless.ai - Documentation: https://meetless.ai/docs - Repository: https://github.com/Meetless/mla - EveryDev.ai: https://www.everydev.ai/tools/meetless-mla