Turn AI coding sessions into team context.
Supermanager turns Claude Code and Codex activity into a private shared workspace with live project feeds, fast summaries, durable team memory, and a read-only MCP for follow-up questions with evidence.
Website · Setup docs · AWS infra
Supermanager is currently a research preview.
- See what your team is doing across Claude Code and Codex without asking everyone for another status update.
- Catch up fast with organization-wide and project-level summaries instead of reading raw transcripts or scrolling terminal history.
- Understand who is driving what through live project feeds, member views, and organization activity in one private workspace.
- Turn repeated work into shared memories and reusable skills so good patterns do not stay trapped in one person's session.
- Bring the same shared context into Claude and Codex through a read-only MCP when you want to ask follow-up questions with evidence.
- Connect repos quickly from the CLI by creating or joining projects, installing the reporting hooks, and syncing shared context back into local agent setups.
- Onboard teams with Google or GitHub sign-in, invite links, and browser-approved CLI logins instead of manual credential wrangling.
- Run against the hosted service or self-host the stack if you want control over the backend and data plane.
- Install the CLI on each machine that runs Claude Code or Codex.
- Run
supermanager loginand approve the device in your browser. - From inside a repo, run
supermanager create projectorsupermanager join <project-id>. - Supermanager installs repo-local hooks so Claude Code and Codex activity starts reporting automatically.
- The web app shows the live feed while the workflow worker turns that activity into summaries, updates, memories, and skills.
supermanager mcp installexposes the same organization context to Claude and Codex over MCP.
curl -fsSL https://supermanager.dev/install.sh | shThe installer downloads the latest GitHub Release for your platform, verifies the published SHA-256 checksum, and installs supermanager into ~/.local/bin by default.
supermanager loginThis starts a device login, opens the approval URL in your browser, and stores the authenticated session locally.
If your account belongs to more than one organization, pick the active one with:
supermanager orgs configureFrom inside the repo you want to track:
supermanager create projectOr, if a teammate already created the project:
supermanager join <project-id>join installs repo-local Claude Code and Codex hooks, stores the repo-scoped API key on the machine, and starts syncing shared context back into your agent setup.
supermanager mcp installThat installs the authenticated Supermanager MCP into your global Claude and Codex config so agents can query accessible projects, summaries, and history directly.
supermanager list
supermanager context sync
supermanager update --check
supermanager leavePublished installs also check once per day for a newer release before normal interactive commands run. Set SUPERMANAGER_AUTO_UPDATE=0 to disable that behavior.
- Bun
1.2.x - Rust toolchain
- PostgreSQL
- Google and GitHub OAuth apps configured for your local API and app origins
CODEX_API_KEYfor embeddings and workflow execution- The
vendor/codexsubmodule initialized
git submodule update --init --recursiveUse .env.example as the base reference. A minimal local setup looks like this:
export DATABASE_URL='postgres://supermanager:password@127.0.0.1:5432/supermanager?sslmode=disable'
export BETTER_AUTH_SECRET='replace-me'
export GOOGLE_CLIENT_ID='replace-me'
export GOOGLE_CLIENT_SECRET='replace-me'
export GITHUB_CLIENT_ID='replace-me'
export GITHUB_CLIENT_SECRET='replace-me'
export CODEX_API_KEY='replace-me'
export SUPERMANAGER_PUBLIC_API_URL='http://127.0.0.1:8787'
export SUPERMANAGER_PUBLIC_APP_URL='http://127.0.0.1:5173'
export SUPERMANAGER_DATA_DIR="$PWD/.supermanager-data"cd packages/server
bun install
bun run devFrom the repo root:
cargo run --manifest-path crates/workflow-agent/Cargo.toml -- \
--database-url "$DATABASE_URL" \
--data-dir "$SUPERMANAGER_DATA_DIR"Optional refresh intervals are available through these environment variables:
SUPERMANAGER_ORGANIZATION_SUMMARY_REFRESH_INTERVAL_SECONDSSUPERMANAGER_PROJECT_SUMMARY_POLL_INTERVAL_SECONDSSUPERMANAGER_PROJECT_MEMORY_EXTRACT_INTERVAL_SECONDSSUPERMANAGER_PROJECT_MEMORY_CONSOLIDATE_INTERVAL_SECONDSSUPERMANAGER_PROJECT_SKILLS_INTERVAL_SECONDSSUPERMANAGER_ORGANIZATION_MEMORY_CONSOLIDATE_INTERVAL_SECONDSSUPERMANAGER_ORGANIZATION_SKILLS_INTERVAL_SECONDS
cd packages/web
bun install
VITE_API_BASE_URL='http://127.0.0.1:8787' bun run devOpen http://127.0.0.1:5173.
From the repo root:
cargo install --path crates/supermanager-cliThen authenticate the local CLI against the local server:
supermanager login --server "http://127.0.0.1:8787"
supermanager create project --server "http://127.0.0.1:8787"crates/
reporter-protocol/ # Shared Rust types for projects, summaries, and hook events
supermanager-cli/ # CLI for auth, repo join/leave, context sync, and MCP install
workflow-agent/ # Workflow worker that maintains summaries, memories, and skills
packages/
common/ # Shared TypeScript protocol types
server/ # Bun + Elysia API, auth, hook ingest, search, and MCP endpoint
web/ # React + Vite marketing site and authenticated workspace
infra/aws/ # Terraform for the AWS deployment
Dockerfile # Backend image build
The Supermanager source in this repository is available under MIT. See LICENSE.
The vendored vendor/codex submodule remains under Apache License 2.0 with
its upstream notices preserved in NOTICE and the files inside that
submodule.