Coordination and automation for agent CLIs.

Sidekar runs alongside Claude Code, Codex, Cursor, Copilot, Gemini CLI, and other agent CLIs.
It is not an orchestrator, harness, or agent OS.
Sidekar is a power packed binary and a skill. That's it.
It adds browser and page automation, data capture, desktop automation, agent coordination, local memory and tasks, background jobs, encrypted secrets, and extension control without replacing the agent you already use.

Install the CLI Browse docs
No MCP Works with your existing agent CLI Also runs as a standalone REPL agent
Browser

One browser surface, end to end

Launch or connect to Chrome, read pages, inspect DOM and accessibility trees, click and type through flows, capture screenshots and PDFs, inspect cookies and network traffic, and work across tabs and frames.

Desktop

macOS UI access

Inspect native windows, take desktop screenshots, click controls, press keys, paste text, launch apps, and finish work that lives outside the browser.

Agent

Coordinate and keep context local

Use the bus, durable memory, task graph, repo pack/tree, and output compaction so agents can collaborate without losing local context.

Jobs + Account

Run in the background, keep secrets local

Schedule recurring prompts and tool runs, monitor tabs for changes, authenticate devices, manage sessions, and store encrypted secrets and TOTP codes.

Web terminal

Check in from anywhere

Sign in to Sidekar and use the relay and web terminal to view active agent sessions remotely without replacing the local PTY as the source of truth.

System + Extension

Own the whole runtime

Manage the daemon, config, updates, event log, skill install, and Chrome extension bridge from the same binary instead of scattering setup across services.

Step 1
Install Sidekar

Run curl -fsSL https://sidekar.dev/install | sh.

Step 2
Authenticate this device (optional)

sidekar device login stores the device token and unlocks account-backed features like the relay tunnel, web terminal, and session management.

Step 3
Install the extension (optional)

Add the Chrome extension for the advanced use-cases. See the extension docs for setup and details.

Step 4
Launch your agent or use Sidekar directly

Wrap your usual agent with sidekar <agent> [args], or run sidekar repl -c <credential> -m <model> for Sidekar's standalone LLM agent mode.

Product pillars

One binary, four main buckets, one supporting layer.

At the product level, Sidekar is browser automation, desktop automation, agent coordination, and background automation. Account, system, and extension commands support those pillars.

Browser

The CLI splits this into Browser, Page, Interact, and Data commands, but it is one capability surface: launch and control Chrome, read pages, interact with real UI, and inspect page state.

Desktop

Use macOS desktop automation for screenshots, app and window inspection, clicks, keypresses, text input, paste, launch, activate, and quit.

Agent

Use the bus, local memory, task graph, agent session history, repo context, and output compaction to keep collaboration local and durable.

Background

Create scheduled jobs with cron, recurring prompt loops with loop, and reactive tab monitoring with monitor.

Supporting layer

Device auth, account sessions, encrypted KV and TOTP, daemon and config management, the local event log, self-update, skill installation, extension control, and sidekar repl support the four main pillars.

Workflows

Work that crosses real boundaries.

Sidekar matters when an agent has to move from code to browser, from browser to desktop, or from one agent to another without losing state.

Ship and verify

One agent edits the code, another opens the actual app in Chrome, and a third watches staging. The bus passes results back without you manually copying context between terminals.

  • sidekar codex builds the change
  • sidekar read or sidekar ext read checks the real page
  • sidekar bus send hands off to QA

Complete login flows

Agents can read credentials from encrypted local storage, generate TOTP codes, and finish browser login flows end to end instead of stopping at “please do this step manually.”

  • Encrypted KV for secrets
  • TOTP generation on demand
  • Real browser and desktop surfaces

Work from local context

Keep project memory, task dependencies, repo context, and compact action output in the same local binary so agents can resume work without rebuilding context from scratch.

  • sidekar memory context loads scoped memory
  • sidekar tasks list --ready shows the next unblocked work
  • sidekar repo pack packs repo context for the model

Watch the background

Keep long-running checks alive after the prompt: watch a dashboard, poll a page, or schedule work and route the result back into the same agent bus.

  • Monitors for reactive alerts
  • Cron and loop for recurring checks
  • Results delivered through the bus
Security and trust

Make the trust model obvious.

Browser automation, desktop automation, local bus delivery, encrypted secrets, and TOTP all stay on your machine. sidekar.dev handles authentication, dashboards, the web terminal, device management, and relay sessions.

What stays local

Browser automation, desktop automation, local bus delivery, encrypted secret storage, and TOTP generation happen on your machine. The extension talks to a local bridge on 127.0.0.1. It does not browse on its own and it does not act until your local Sidekar process tells it to.

Remote features exist for authentication, dashboards, and relay sessions. They are a separate layer, and this site spells out that boundary instead of hiding it in docs.

Install sidekar Open docs