The connector vision

Meetless Agent separates two things that are usually tangled together: the governance layer and the surface the agent runs on.

• The governance layer is the model: governed memory, the trust and promotion lifecycle, injection, ask, capture, contradiction detection, and enforcement. It is agent-agnostic. None of it is specific to any one tool.
• A connector wires that layer into a specific place an agent works: its prompt path, its tool calls, its produced artifacts.

Claude Code is connector number one Available today

The first connector is Claude Code. It wires the governance layer into a local Claude Code setup through three surfaces working together:

• Capture hooks that observe prompts, decisions, tool calls, and produced documents.
• An MCP server (mla mcp) that lets the agent query governed memory from inside a session.
• A /mla skill that drives onboarding and review flows.

Everything in How it works is taught agent-agnostic, then landed on Claude Code in The Claude Code connector.

OpenAI Codex is connector number two Available today

The Codex connector wires the same governance layer into Codex CLI through the same three jobs, using Codex’s own primitives:

• Hooks registered in $CODEX_HOME/hooks.json by mla codex install, for floor injection on UserPromptSubmit and enforcement on PreToolUse.
• An MCP server (the same mla mcp), registered as a Codex plugin.
• The same neutral core, so mla activate binds a repo without a Codex-specific branch.

Codex support has two independent halves, hooks and MCP, and neither implies the other, so mla doctor reports it as three separate checks. Two things we do not soften: Codex hooks fail open until you grant trust with /hooks, and enforcement is advisory by default. The full contract, including what enforcement does and does not do today, is in the Codex connector README. A dedicated Codex section on this site is still to come; until it exists we link to the source of truth rather than a thinner summary of it.

More connectors are the direction Planned

Because the governance layer does not depend on any one agent, the same governed memory and the same trust model can front others. Beyond Claude Code and Codex, additional connectors are planned, not shipped.

Note

We describe unshipped connectors as direction, never as present tense. Codex was described that way here until it actually shipped, at which point this page grew a sibling section and nothing else in the model changed. That is the test the framing was built to pass.

Why this framing matters to you

Adopting Meetless Agent for one agent is not adopting a single-agent product. You are adopting a governance model, with surfaces attached. The concepts you learn, the corpus you build, and the trust decisions you make carry forward across agents: a decision your Claude Code session got approved governs your Codex sessions too, because the decision logic lives in the core rather than in either connector. That is the whole point of teaching the model once and landing it on one connector at a time.