yan5xu/codexloom

Turn Codex threads into an organization of long-lived domain agents.

CodexLoom – Persistent AI‑Agent Workspaces

What it is – CodexLoom is a self‑hosted platform that extends the Codex chat‑agent runtime with durable identities, profiles, and organizational structures. It lets a single Domain Agent keep a stable ID, name, profile and primary thread that lives across many tasks, while other agents communicate with it through bounded messages, topics, and managed artifacts. The system also provides “Interface Agents” that expose these internal agents to external chat tools such as Feishu (Lark), Slack, and Parall.

Key capabilities

  • Long‑lived agents – stable ID, editable profile, persistent thread, and configurable model settings.
  • Multi‑surface access – the same thread can be used from Codex Desktop, Mobile, the CodexLoom Web UI, or the loom CLI with live sync.
  • Agent‑to‑agent messaging – queue, send, reply, and track delivery state while preserving full history.
  • Bounded coordination – Topics for scoped collaboration, “Needs You” prompts for human decisions, and managed Artifact hand‑offs.
  • Team organization – draggable maps for Organization (responsibility hierarchy), Collaboration (cross‑domain links), Activity (evidence), and a Directory of all agents.
  • Governed external delivery – Interface Agents map agents to external IM conversations, enforce conversation‑membership policies, and keep identity/authorization boundaries.
  • Operational tools – schedules, external triggers, runtime inspection, on‑demand backups, graceful restarts, and usage dashboards (tokens, context, cache, model).

Who it’s for – Advanced solo users or one‑person company owners who want several specialized agents to run continuously, integrate them into the chat tools they already use, and keep the agents’ knowledge across tasks. Enterprise multi‑tenant features are not the primary focus yet.

Quick start

# Prerequisite: install the `codex` CLI and sign in with a ChatGPT account
make release               # builds the binary
./bin/codex-loom           # starts the local server (http://localhost:4870)

You can then create an agent via the Web UI or the CLI, set its profile, and send the first message:

./bin/loom agent create research --cwd /path/to/repo
./bin/loom profile set research \
  --identity "Long‑term researcher" \
  --domain "Research the product ecosystem" \
  --scope "Answer questions, preserve conclusions"
./bin/loom thread send research "Establish a baseline for this domain"

Community & docs – A Feishu community is provided for discussion. Full documentation (owner guides, API specs, integration notes, CLI reference, architecture handbook, etc.) lives under docs/.

License & status – Source‑available under the Elastic License 2.0 (not OSI‑approved open source). The project is local‑first, self‑hosted, and under active development; remote features depend on experimental Codex APIs that may change.

All information above is taken directly from the repository’s README; no additional claims are made.

Related