LilithGames/a2wave

npm i -g a2wave && a2wave setup
a2wave setup --yes --with-postgres --dir "$HOME/a2wave-pg" --port 3512
docker compose up -d --build

Usage

• •After setup, visit http://localhost:3502, create an Agent, bind a model provider, and publish it (in-app manual at /wiki).
• •Describe an Agent in natural language; a2wave orchestrates the underlying CLI — it does not bundle inference or a sandbox runtime.
• •Extend Agents by composition with Skills and MCP servers (stdio / SSE / HTTP / proxy groups) instead of forking.
• •Set ADMIN_PASSWORD in .env, or the first person to reach the setup page claims the admin account.

Key features

• ✓Bring-your-own agent CLIs installed on demand from a pinned, checksum-verified lockfile.
• ✓Publish one Agent to HTTP API, Feishu, Slack, Discord, A2A, schedules, GitLab/GitHub triggers and a first-party chat page.
• ✓Agent-to-agent calls over A2A, including Agents hosted outside your deployment.
• ✓Persistent per-Agent memory (keyword/vector/hybrid), built-in evaluation with frozen provider snapshots, and Git/Perforce worktree isolation.
• ✓Enterprise auth: OIDC/SAML SSO, per-Agent owner/editor/viewer permissions, rate limiting and an audit entry behind every write.

Best for

Internal teams that already trust an agent CLI and need to share it across Feishu/Slack/Discord with access control, audit trails and delivery.

Caveats

• ⚠Built for trusted internal teams; exposing it to untrusted users or running untrusted Agent configs is out of scope — add your own isolation layer.
• ⚠PostgreSQL backend is experimental (passes tests but no production soak; no SQLite→Postgres migration path).
• ⚠Feishu channel supports feishu.cn apps only; Lark international (larksuite.com) is not configurable yet.
• ⚠macOS needs A2WAVE_WORKSPACE_DIR and pinned UID/GID in .env due to Docker Desktop sharing limits.
• ⚠Apache-2.0 licensed.