What is Cezar?

Cezar is a self-hosted orchestration tool for AI coding assistants. It runs several agents — like Claude Code, Codex, and OpenCode — concurrently on your machine or a remote server, with each task isolated in its own git worktree. A browser-based dashboard shows live output, tool calls, and token usage. You define pipelines as YAML workflows that can include shell checks and automatic retries. Recurring or event-driven runs can be triggered from GitHub, Jira, or Linear. No database is required; everything persists as plain files. Installation options include a one-line npx command, a VPS installer that sets up nginx with HTTPS and basic auth, and native desktop apps for major operating systems. The codebase is MIT-licensed and actively maintained.

Should you self-host Cezar?

Cezar scores 55/100 on selfhost.directory (how scoring works) and ranks #65 of 209 in AI Agents. It has 859 GitHub stars.

It shipped 10 stable releases in the past year. The latest, v0.16.3, came out on 24 October 2026. New versions land roughly every 1 days.

Setup is moderate: expect some configuration beyond a single command. It is MIT-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is Orca (91/100, 108 446 stars). It scores higher overall, so compare the two before you pick. See Cezar vs Orca side by side.

Features

Automations (schedules & event triggers)
Autonomous mode
Desktop app (beta)Beta
Live streaming cockpit
Memory-aware runs
Multi-project cockpit
Parallel agent orchestration
Parallel variants (×2/×3)
Review gate
Skills System
Task dispatch (child subtasks)
Usage and cost tracking
Workflow engine with YAML
Workspace dashboard

How to configure Cezar

CEZ_AGENT_ENV_FULLEscape hatch: give spawned agents the full host environment
Escape hatch: give spawned agents the full host environment
values: 1
CEZ_AGENT_MODELS_LOCKEDGlobally lock each runner to the model configured in its native settings
Globally lock each runner to the model configured in its native settings
values: 1
CEZ_AGENT_TMPDIRStop giving each task its own temp directory (on by default, only exact 0 disables)
Stop giving each task its own temp directory (on by default, only exact 0 disables)
values: 0
CEZ_APPROVAL_GATESelect Claude's acceptEdits mode (cezar cannot answer permission prompts)
Select Claude's acceptEdits mode (cezar cannot answer permission prompts)
values: 1
CEZ_AUTOMATIONSTurn automations off (only exact value 0 opts out)
Turn automations off (only exact value 0 opts out)
values: 0
CEZ_AUTONAMEDisable ALL LLM task naming (creation + live)
Disable ALL LLM task naming (creation + live)
values: 0
CEZ_AUTONOMOUS_DEFAULTSeed the New Task Autonomous default
Seed the New Task Autonomous default
values: 0, 1
CEZ_AUTOSAVERe-enable periodic (90s) autosave commit in task worktrees (off by default)
Re-enable periodic (90s) autosave commit in task worktrees (off by default)
values: 1
CEZ_BROWSE_ROOT~/Default root for Add project → Open local folder…
Default root for Add project → Open local folder…
CEZ_CLAUDE_BINOverride which claude binary is used
Override which claude binary is used
values: /path/to/claude
CEZ_CODEX_BINOverride which codex binary is used
Override which codex binary is used
values: /path/to/codex
CEZ_COPILOT_BINOverride which GitHub Copilot CLI binary is used
Override which GitHub Copilot CLI binary is used
values: /path/to/copilot
CEZ_CURSOR_AGENT_BINOverride which Cursor Agent CLI binary is used
Override which Cursor Agent CLI binary is used
values: /path/to/cursor-agent
CEZ_DESIGN_MODETurn off Design Mode in a task's Preview tab
Turn off Design Mode in a task's Preview tab
values: 0
CEZ_DISABLE_REPO_LOCKDangerous escape hatch: allow runs without Cezar's repository-root lease
Dangerous escape hatch: allow runs without Cezar's repository-root lease
values: 1
CEZ_DISPATCHTurn OFF task dispatch (on by default, only exact value 0 turns it off)
Turn OFF task dispatch (on by default, only exact value 0 turns it off)
values: 0
CEZ_DRY_RUNUse the bundled mock agent instead of real CLIs for offline demos and development
Use the bundled mock agent instead of real CLIs for offline demos and development
values: 1
CEZ_ENV_PASSTHROUGHForward extra host env vars to spawned agents (comma-separated)
Forward extra host env vars to spawned agents (comma-separated)
values: A,B
CEZ_FOLLOWUPSTurn on the global follow-up Inbox (agents leave follow-ups in todos.json)
Turn on the global follow-up Inbox (agents leave follow-ups in todos.json)
values: 1
CEZ_HIDE_COSTHide backend-reported monetary cost throughout the browser cockpit
Hide backend-reported monetary cost throughout the browser cockpit
values: 1
CEZ_HIDE_TOKEN_METRICSLegacy master switch that hides both token usage and cost
Legacy master switch that hides both token usage and cost
values: 1
CEZ_HIDE_TOKEN_USAGEHide raw input/output token counts throughout the browser cockpit
Hide raw input/output token counts throughout the browser cockpit
values: 1
CEZ_INSTANCE_IDInternal server-install identity set automatically in generated systemd/launchd services
Internal server-install identity set automatically in generated systemd/launchd services
CEZ_JUNIE_BINOverride which junie binary is used
Override which junie binary is used
values: /path/to/junie
CEZ_NO_BANNERSilence every promo: skills banner, star-ask line, sidebar chip, star-ask dialog
Silence every promo: skills banner, star-ask line, sidebar chip, star-ask dialog
values: 1
CEZ_OPENCODE_BINOverride which opencode binary is used
Override which opencode binary is used
values: /path/to/opencode
CEZ_PI_BINOverride which pi binary is used
Override which pi binary is used
values: /path/to/pi
CEZ_PROJECTS_DIR~/cezar/projectsDefault destination for Clone from GitHub
Default destination for Clone from GitHub
CEZ_REDACT_SECRETSDisable scrubbing of credential values/token shapes from on-disk state (on by default)
Disable scrubbing of credential values/token shapes from on-disk state (on by default)
values: 0
CEZ_REMOTE_AGENT_ACCOUNTSAllow Settings → Agent accounts in a remote/self-hosted cockpit (requires authenticated perimeter)
Allow Settings → Agent accounts in a remote/self-hosted cockpit (requires authenticated perimeter)
values: 1
CEZ_REVIEW_GATETurn ON the optional diff-first review gate (off by default, only 1 enables)
Turn ON the optional diff-first review gate (off by default, only 1 enables)
values: 1
CEZ_SECRETS_KEYCHAINKeep the secret store's data key out of the OS keychain and use the 0600 key file only
Keep the secret store's data key out of the OS keychain and use the 0600 key file only
values: 0
CEZ_SINGLE_PROJECTOpt into a launch-project-only cockpit (only exact value 1 enables)
Opt into a launch-project-only cockpit (only exact value 1 enables)
values: 1
CEZ_SKILLS_AUTO_UPDATEDisable automatic checks and updates for upstream-CLI-tracked Open Mercato skill installations
Disable automatic checks and updates for upstream-CLI-tracked Open Mercato skill installations
values: 0
CEZ_SUPERVISEDA supervisor relaunches cezar after an update (desktop shell sets this)
A supervisor relaunches cezar after an update (desktop shell sets this)
values: 1
CEZ_TASK_PROCESS_REAPTurn OFF leftover-process reaping (on by default, only exact value 0 turns it off)
Turn OFF leftover-process reaping (on by default, only exact value 0 turns it off)
values: 0
CEZ_TELEMETRYBlock usage telemetry even after opt-in (0) or lift CI/test-runner block (1)
Block usage telemetry even after opt-in (0) or lift CI/test-runner block (1)
values: 0, 1
CEZ_TERMINALEnable/disable the task view's interactive terminal (on for local, off for hosted)
Enable/disable the task view's interactive terminal (on for local, off for hosted)
values: 1, 0
CEZ_TITLE_UPDATESTurn off the live task-title refresh
Turn off the live task-title refresh
values: 0
CEZ_UPDATE_CHANNELstableRelease channel the self-updater follows
Release channel the self-updater follows
values: stable, nightly, development
CEZ_WORKFLOW_WEBHOOKSLet a workflow graph's notify.webhook node POST to the URL it names
Let a workflow graph's notify.webhook node POST to the URL it names
values: 1
CEZ_WORKTREE_DEFAULTSeed the New Task Worktree default
Seed the New Task Worktree default
values: 0, 1
VITE_CEZ_API_BASEBuild time only: API base when cockpit bundle is deployed apart from the service
Build time only: API base when cockpit bundle is deployed apart from the service
values: http://localhost:4321

FAQ

Can I try it without logging in to an agent?

Yes. CEZ_DRY_RUN=1 npx cezar-run starts the cockpit with a built-in mock agent, so you can look around first.

Who builds it?

Cezar was created by Patryk Lewczuk at Open Mercato, the open source CRM/ERP framework company founded by Piotr Karwatka. Development happens in the open at github.com/open-mercato/cezar.

What does it cost?

Cezar is free, MIT-licensed open source. You pay only for the agent plans you already have, and the Dashboard shows reported usage and cost per task and project.

What can start a run without me?

Automations: a schedule, a GitHub poll, a PR review, or a Jira or Linear event. With Autonomous on, the run never stops to ask and finishes with the PR ready for review.

Will it ship broken code?

Not without you. Workflow checks like npm test gate every step (exit 0 means pass) and failures go back to the agent to retry. Every run with changes waits at a review gate, and nothing merges on its own.

How many agents can run at once?

As many as your machine can handle. Set a parallel limit for the workspace or per project and the rest waits in a queue that survives restarts. Running tasks can dispatch subtasks of their own, and a memory ceiling pauses heavy runs so the queue keeps moving.

How does it keep working when my laptop is closed?

Run npx cezar-run server-install --platform ubuntu-vps. It sets up HTTPS, a login and a system service, so the cockpit is reachable from anywhere, phone included. Alternatively, try it on Open Mercato Cloud.

Which agents can it run?

Claude Code, Codex, OpenCode and pi today, with more added as new agent CLIs come out. Pick one per task or per workflow step, and switch a task's runner, model or account mid-thread without losing the conversation.

Do I need API keys?

No. Cezar drives the agent CLIs you are already logged into, so runs use the subscriptions you already have. There is no cezar account and no database either.

What is cezar?

An open-source orchestrator for AI coding agents that runs Claude Code, Codex, OpenCode and other coding agents in parallel, each task in its own git worktree, on your laptop or 24/7 on a server, with a web cockpit to watch, steer and review everything.