# AgentBridge — Local bridge connecting Claude Code and Codex as peers: cross-review, task splits, quota relay. > Source: The Agents Index — https://theagentsindex.com/agentbridge (structured, researched, re-verified) > Facts last verified: 2026-09-02 AgentBridge is a local CLI and daemon that connects a Claude Code session and a Codex TUI as peers in one working session. Either side can push a message into the other's running turn, so cross-review and task handoff need no human relay. | Fact | Value | | --- | --- | | Website | https://raysonmeng.github.io/agent-bridge | | Free tier | Yes | The whole tool is free under MIT with no vendor-imposed limits. Practical requirements are Bun as the runtime plus your own Claude Code and Codex CLIs and logins; macOS and Linux only, Windows is not officially supported. ## What it does **Writes code** · No — AgentBridge relays messages between two coding agents; the code is authored by Claude Code and Codex, not by the bridge. > A bridge that forwards messages between an MCP channel and the Codex app-server protocol > — https://github.com/raysonmeng/agent-bridge · README, 'What this project is / is not' — 'This project is' list, second bullet **Debugs and fixes** · No — Its only diagnostic command inspects the bridge's own health, not your program; no failure diagnosis or fix landing. > Read-only diagnosis: env, daemon health/readiness, build drift, artifact alignment, TUI attachment, logs > — https://github.com/raysonmeng/agent-bridge · README, CLI Reference table, 'abg doctor [--json]' row **Writes tests** · No — No test generation. The bridge never sees file changes or command execution, and the test count advertised on the homepage is AgentBridge's own suite, not a product feature. > Only forwards agentMessage items, not intermediate commandExecution, fileChange, or similar events > — https://github.com/raysonmeng/agent-bridge · README, Current Limitations, first bullet **Reviews code** · Partly — It routes one agent's review of the other's diff into a running turn; the reviewing itself is done by Claude or Codex, not by AgentBridge. > Codex implements; Claude reviews the diff in real time inside the same session and pushes change requests straight back into Codex's thread. > — https://raysonmeng.github.io/agent-bridge/ · Three workflows section, 'Cross-review, inside the session' card **Whole-codebase work** · No — The bridge carries messages, not edits. Repository-wide work is executed by the two connected agents, and scope is one pair per project directory. > AgentBridge only relays messages between them on your machine; raw work never leaves the box. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' **Ships it** · No — No deploy or release step: the command set (init, claude, codex, resume, pairs, doctor, budget, logs, kill) is local development plumbing only. > A local developer tool for connecting Claude Code and Codex in one workflow > — https://github.com/raysonmeng/agent-bridge · README, 'What this project is / is not' list and the CLI Reference table **Done-for-you service** · No — Software you install and run yourself: a global npm package plus a local daemon. There is no delivery team and no managed service. > A hosted service or multi-tenant system > — https://github.com/raysonmeng/agent-bridge · README, 'What this project is / is not' — 'This project is not' list, first bullet ## How much it does unattended **Runs autonomously** · Partly — Both agents launch with permission prompts disabled so a pair can run unattended, but the vendor's own framing is that you steer while the agents coordinate. > By default abg claude launches with --dangerously-skip-permissions and abg codex with --yolo. This is deliberate: an unattended agent pair can't stop to ask you for each permission. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Is it safe? What does --safe do?' **Multi-agent** · Yes — This is the entire product: two agents from different providers held live as persistent peers, each able to push into the other's running turn. > AgentBridge is a local bridge that lets Claude Code and Codex collaborate in real time inside one working session > — https://raysonmeng.github.io/agent-bridge/ · Hero section, paragraph under the headline **Agent permissions** · Partly — Two states only: max-permission by default, or --safe to restore each CLI's normal prompts. The vendor states it is not a security boundary. > Add --safe (or set AGENTBRIDGE_SAFE=1) to launch with normal prompts. AgentBridge is a local dev tool, not a hardened boundary between untrusted tools. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Is it safe? What does --safe do?' ## Which models it runs on **Claude** · Yes — Claude Code is one of the two supported sides; it is started as 'abg claude' with the AgentBridge channel enabled, using your own Claude login. > Today the shipping in-session integration connects Claude Code and Codex. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Which agents does it support?' **GPT** · Yes — OpenAI's Codex CLI is the other side of the pair, started as 'abg codex' in a second terminal against the same bridge. > Today the shipping in-session integration connects Claude Code and Codex. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Which agents does it support?' **Open models** · No — The bridge selects no model at all, and the closed list of wired agents is the two proprietary CLIs; further adapters are a roadmap vote, not shipped. > Today the shipping in-session integration connects Claude Code and Codex. Which agent comes next is a public vote. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Which agents does it support?' **Model choice** · Partly — AgentBridge picks no model and no boss role; whatever each CLI is configured to run is what runs. The choice is confined to the two supported CLIs. > You use your existing Claude Code and Codex logins; AgentBridge only relays messages between them on your machine > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' ## Where you use it **In your editor** · No — Both surfaces are terminal: 'abg claude' starts Claude Code with the channel and 'abg codex' starts the Codex TUI. No IDE or editor integration is offered. > Start Codex TUI connected to AgentBridge daemon. > — https://github.com/raysonmeng/agent-bridge · README, CLI Reference table, 'abg codex [args...]' row **On the command line** · Yes — The product is a CLI plus a background daemon; every capability is reached through terminal commands under 'agentbridge' or the 'abg' alias. > All commands work with both agentbridge and the short alias abg. > — https://github.com/raysonmeng/agent-bridge · README, CLI Reference section, note above the command table **In your pipeline** · No — Nothing acts on a pull request or in CI. The enumerated command set is local-only, with no webhook, PR or pipeline surface. > A local developer tool for connecting Claude Code and Codex in one workflow > — https://github.com/raysonmeng/agent-bridge · README, 'What this project is / is not' — 'This project is' list, first bullet **In a browser** · No — No hosted dashboard or web app; the GitHub Pages site is marketing and every runtime component runs on your own machine. > AgentBridge is a local tool with zero runtime dependencies and no hosted service. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' ## Whose machine it runs on **Self-hosted** · Yes — Local-only by design: an npm-installed CLI and a background daemon on your machine, with no hosted component involved. > AgentBridge only relays messages between them on your machine; raw work never leaves the box. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' **Open source** · Yes — MIT licensed, with the whole implementation public in the vendor's GitHub repository. > Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction > — https://github.com/raysonmeng/agent-bridge/blob/master/LICENSE · LICENSE, MIT grant paragraph ## What it costs to run **How it meters** · No — Nothing is metered or billed by the vendor: no seats, tokens, credits or tasks. Your only costs are the Claude Code and Codex subscriptions you already pay for. > No. AgentBridge is a local tool with zero runtime dependencies and no hosted service. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' **Free tier** · Yes — Free of charge under the MIT licence; the whole tool is a public npm package you install and run yourself, with no call and no signup. > Permission is hereby granted, free of charge, to any person obtaining a copy of this software > — https://github.com/raysonmeng/agent-bridge/blob/master/LICENSE · LICENSE, MIT grant paragraph **API access** · Partly — No product API for external callers. The programmable surfaces are the MCP tools the plugin exposes to Claude, the local control WebSocket, and --json output from doctor and budget. > Claude calls the reply tool -> bridge.ts -> control WS -> daemon.ts -> turn/start injects into the Codex thread > — https://github.com/raysonmeng/agent-bridge · README, Architecture / Data flow table, 'Claude -> Codex' row **MCP server** · Yes — MCP is the transport by which the paired agent addresses it: the Claude Code plugin runs an MCP channel whose reply and get_messages tools carry the traffic. There is no remote MCP endpoint. > the daemon intercepts Codex output and pushes it to Claude as channel notifications; Claude replies via the reply MCP tool > — https://github.com/raysonmeng/agent-bridge · README, Features list, 'Bidirectional Claude ↔ Codex messaging' bullet **Bring your own key** · Yes — Entirely your own inference: it drives the Claude Code and Codex sessions already installed on your machine and needs no account or key of its own. > You use your existing Claude Code and Codex logins; AgentBridge only relays messages between them on your machine > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' ## Buying it for a team **Pooled budget** · Partly — No seat balances exist, but the daemon does track both agents' subscription quota windows jointly and reports drift between the two sides. > The daemon polls both agents' account-level 5h and weekly quota. abg budget --json prints the live snapshot: both windows, drift between the two sides, and pause state. > — https://raysonmeng.github.io/agent-bridge/ · Quota relay section, 'Both windows in one glance' card **Admin controls** · Partly — Constraints exist but are per-machine developer settings (--safe, AGENTBRIDGE_* environment variables and .agentbridge/config.json), with no admin role and no centrally enforced policy. > Set AGENTBRIDGE_FILTER_MODE=full (or the config equivalent) when you do want the unfiltered stream. > — https://github.com/raysonmeng/agent-bridge · README, 'Context handling' section, closing paragraph; see also Configuration → Environment Variables table **Audit log** · Partly — Local daemon and Codex-wrapper logs only, tailed with 'abg logs' and stored in a platform state directory. There is no account-level audit trail, because there are no accounts. > Tail this pair's daemon log (or the Codex wrapper log with --codex); -f follows, -n N sets the line count (default 100) > — https://github.com/raysonmeng/agent-bridge · README, CLI Reference table, 'abg logs' row **Single sign-on** · No — There is no identity layer to federate: AgentBridge holds no accounts, and you authenticate to Claude Code and Codex separately as yourself. > No. AgentBridge is a local tool with zero runtime dependencies and no hosted service. You use your existing Claude Code and Codex logins > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' ## What happens to your code **Opt out of training** · Partly — The bridge itself sends nothing off the machine, so it has nothing to opt out of; what Claude Code and Codex do with your code is governed by those vendors, not by AgentBridge. > AgentBridge only relays messages between them on your machine; raw work never leaves the box. > — https://raysonmeng.github.io/agent-bridge/ · FAQ, 'Do I need an account or API keys for AgentBridge itself?' **Data residency** · Yes — Residency is wherever you run it: every bridge connection is loopback-only and runtime state lives in a local per-platform directory, with no vendor-side store. > All connections are on 127.0.0.1 — no external network exposure by default. > — https://github.com/raysonmeng/agent-bridge/blob/master/SECURITY.md · Security Considerations, 'Local WebSocket connections' bullet **Getting out** · Yes — All state is plain local files you already own: .agentbridge/config.json, checkpoints and logs in a platform state directory. There are no credits to expire. > Contents: daemon.pid, status.json, agentbridge.log, killed (sentinel), startup.lock > — https://github.com/raysonmeng/agent-bridge · README, State Directory section, contents line; see also the Project Config table ## Provenance 3 of 4 sources behind the facts above are independent of the vendor. ## Coverage elsewhere None recorded. Every fact above comes from the vendor and is not independently corroborated yet. ## What changed - 2026-09-02 — Capabilities · Commercial terms · Free tier - 2026-09-01 — Capabilities · Commercial terms · Free tier ## Correcting this record Anything marked "Not established" is a question we have not answered — it is not a No. If you represent AgentBridge, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth. Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.