Sikemux
A whole team of AI coders in one window, so you can see who is working, who is stuck and who needs you.
- Source repo
- nodelike/sikemux
- Stars
- ★ 36
- Last updated
- today
- License
- MIT
- Primary language
- Rust
- FA score
- 49/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude CodeOpenAI API (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Run Claude Code and Codex on different tasks in the same repository, compare progress in one window and take over whichever gets stuck.
- Not a fit if
- Users on Windows or Intel Macs
- Solo users who only want one chat assistant, not multi-agent orchestration
- Website
- sikemux.com ↗ set on GitHub
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-20latest release mobile-v0.2.1 (2026-11-20)
- Source review
- 49/100 · Major gaps
What does this agent do, and when should you use it?
Sikemux is a native desktop app written in Rust (about 30 MB, not Electron) that places multiple AI coding agents — Claude Code, Codex, OpenCode, Hermes, Pi and Muse — side by side in a single window, each with its own browser tabs. Around those agents it wraps a code editor, terminals, Git and search with 450+ themes, so multi-agent work, code changes and version control happen in one surface. You can @ an agent, # an issue, ! a PR, + a file, & something from the stash, or hand over a terminal selection or log; any chat can get its own worktree. Tools such as GitHub, GitLab, Jira, Linear, Slack, Vercel, AWS and your databases are signed in once and shared by every agent. Quitting does not stop terminals, agents or processes — they return where you left them — and chats and terminals can run on any host on your account, including turning an SSH host into one in a click. It is free under MIT for Apple Silicon Macs (macOS 11+) and for x86_64 and arm64 Linux.
On launch, Sikemux keeps an independent session and browser tab per AI coding agent (Claude Code, Codex, OpenCode, Hermes, Pi, Muse) inside one native GPUI-rendered window. Users pass context with @ for an agent, # for an issue, ! for a PR, + for a file, & for a stash entry, or by handing over a terminal selection or log; each chat can be bound to its own Git worktree so parallel edits do not collide. A code editor, terminals, Git panel and search sit alongside the sessions so changes can be inspected and committed directly. GitHub, GitLab, Jira, Linear, Slack, Vercel, AWS and databases are authorized once and then callable by every agent. Terminals and agents keep running after the app is closed and are restored on reopen, and chats and terminals can be dispatched to any host on the account, with SSH hosts onboarded in one click. The install script also offers a --host mode that installs only the core on a headless server.
- Run Claude Code and Codex on different tasks in the same repository, compare progress in one window and take over whichever gets stuck.
- Give one agent its own worktree to fix a bug while another writes tests on the main branch without overwriting each other.
- Link a Jira or Linear ticket with # and a PR with ! so requirements, code and review live in one agent session.
- Deploy the core with --host on a remote Linux server and drive its terminals and agents from a local Mac window.
- Hand a stack trace or terminal selection straight to an agent instead of copy-pasting between windows.
- Sign in once to AWS, Vercel or a database so different agents can check deployment state and data while debugging.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
curl -fsSL https://sikemux.com/install | shbrew install --cask nodelike/tap/sikemuxcurl -fsSL https://sikemux.com/install | sh -s -- --host
- Runtime
- Node.js 22+
- Install notes cover
- macOS · Linux
How do you install or deploy this agent?
On macOS or Linux, run the official install script:
curl -fsSL https://sikemux.com/install | shOn a Mac you can also use Homebrew:
brew install --cask nodelike/tap/sikemuxFor a headless server, install just the core:
curl -fsSL https://sikemux.com/install | sh -s -- --hostA .dmg, .deb, Arch package and AppImage are also on the Releases page. Requirements are macOS 11+ on Apple Silicon, or Linux on x86_64 and arm64. Building from source needs Rust, Node.js 22+ and pnpm:
git clone [email protected]:nodelike/sikemux.git && cd sikemux
pnpm install
make devHow do you use this agent?
Open Sikemux after installing; it runs the Claude Code, Codex and other subscriptions you already pay for, and needs no account on your own machine. Create an agent session in the window and route context with the prefix keys: @ for an agent, # for an issue, ! for a PR, + for a file, & for a stash entry, or select terminal output and hand it over. For parallel edits, assign the chat its own worktree. The first time you use GitHub, GitLab, Jira, Linear, Slack, Vercel, AWS or a database, sign in once and every agent shares that authorization. To attach a remote machine, add an SSH host in settings, or deploy the core with --host and connect from the client. For a development build, run make dev, which installs as a separate Sikemux Dev app next to the regular one.
What are this agent's strengths and limitations?
- Native Rust/GPUI app at roughly 30 MB rather than Electron, so startup and resource shape differ from browser-based IDEs.
- Runs Claude Code, Codex, OpenCode, Hermes, Pi and Muse together in one window instead of locking you to one model vendor.
- Each chat can be bound to its own Git worktree, isolating parallel agent edits by branch.
- Terminals, agents and processes survive quitting the app and come back where you left them, so long tasks are not cut off.
- Tool integrations (GitHub, GitLab, Jira, Linear, Slack, Vercel, AWS, databases) are signed in once and shared across all agents.
- MIT-licensed and free to use, on Apple Silicon Macs and x86_64/arm64 Linux.
- Only Apple Silicon Macs (macOS 11+) and Linux are supported; Intel Macs and Windows are not.
- You must bring your own paid Claude Code, Codex or similar subscription, or use free models, so real usage has model costs.
- Building from source requires Rust, Node.js 22+ and pnpm together — a medium-effort developer setup.
- Multi-agent sessions, worktrees and remote-host dispatch add real conceptual overhead; single-agent users gain little for the complexity.
- The repository carries no AGENTS.md-style agent configuration spec and says little about offline operation or where agent data is stored.
How does this agent compare with similar options?
The README contrasts Sikemux with single-assistant tools and with Electron apps but names no specific competitor, so no named comparison is made.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Sikemux This agent | 49 · Major gaps | Desktop appFree + model costs | ★ 36 | today | Rust | Codex · Claude Code |
| Yardsort | 83 · Good | Desktop appFree | ★ 14 | 1d ago | Rust | Codex · Claude Code |
| Diri | 75 · Good | Desktop appFree + model costs | ★ 499 | 1d ago | Rust | Codex · Claude Code |
| Helmor Local Agent Workbench | 60 · Some gaps | Desktop appFree + model costs | ★ 1.3k | 3mo ago | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Trust is thin overall. Least privilege: AGENTS.md tells agents to commit directly to main and use git_commit/agent_spawn, and SKILL.md routes agents to jira/aws/vercel tools carrying the user's sign-in, yet no explicit tool permission scopes or resource boundaries are shown — deducted. User confirmation: project-setup mentions the person is asked once to trust new steps, but the curl | sh installer and the auto-publishing stable release flow weaken confirmation — deducted. Data-flow transparency: README says tools are signed in once and usable by every agent, but does not explain credential storage or per-agent isolation — deducted. Sensitive data handling: server code handles STASH_KEY, Clerk, Paddle, APNs/FCM secrets, but the sample lacks a full key-rotation or least-exposure policy — deducted. Dependency security: CI runs cargo-audit and pnpm deps:audit and verifies downloaded shellcheck/cargo-audit with SHA256, which is good practice, but no audit results or vulnerability remediation records are shown — 2. External effects: agents can run git_commit, agent_spawn, browser and simulator actions with wide blast radius, and no unified guardrail for destructive operations is shown — deducted. Rollback: roll-back-update/withdraw-update commands and worktree.onRemove exist, but undo for an agent's own actions is unclear — deducted. Source attribution: LICENSE and README name nodelike, but publisher identity is unverified and third-party code/model attribution is not described — deducted.
Self-consistency: Cargo.toml version 0.6.1-nightly.22 matches package.json, and AGENTS.md and CI agree on generated-file checks — 2. Dependency availability: README states Rust, Node 22+ and pnpm prerequisites, and Cargo.toml patches to a pinned sikemux/gpui rev, so sources are explicit, though the fork's availability cannot be statically confirmed — 2. Failure messages: server/api/src/main.ts gives clear logs and exit codes for UpdateRefused, usage errors and missing config, and server.ts warns per missing key — 2.
Audience and scenarios: README clearly targets developers on Mac/Linux using Claude Code, Codex and similar — 2. Capability boundaries: SKILL.md lists tool families but does not systematically state what agents must not do or which actions need human approval — deducted. Trigger precision: project-setup's description gives load conditions, but the sikemux skill description is broad ('before any other MCP, CLI') and may over-trigger — deducted. Environment fit: README covers Apple Silicon, Linux x86_64/arm64 and headless install — 2.
Information architecture: README, AGENTS.md, CONTRIBUTING reference, docs-site and per-folder AGENTS.md form a layered structure — 2. Install notes: README gives curl, Homebrew and headless install plus build steps — 2. Naming stability: AGENTS.md states released methods are never removed or renamed and defines PROTOCOL_VERSION rules — 2. Examples and FAQ: README has screenshots and a feature list but no FAQ or troubleshooting — deducted. Known limitations: README notes Apple Silicon Mac and Linux only, but no feature limitations or known issues are listed — deducted. License: LICENSE is full MIT text, consistent with README and badge — 3. Versioning/changelog: version numbers and a nightly badge exist, but no CHANGELOG in the sample — deducted. Maintenance responsibility: AGENTS.md declares 'you are the owner and long-term maintainer', but publisher identity is unverified and no maintainer contact or response commitment is shown — deducted.
Output usability: SKILL.md gives concrete command formats, argument passing and error handling that agents can use directly — 2. Marginal value: side-by-side multi-agent panes, worktree isolation and setup_check dual-checkout verification add value over generic single-agent tools — 2. Cost-benefit: README claims a ~30MB native app, free and MIT, but parallel agents consume substantial model quota and machine resources, and the sample gives no resource or cost-control guidance — deducted.
Claim traceability: most README feature claims have no corresponding code or test references in the sample — deducted. Cross-source corroboration: version numbers match between Cargo.toml and package.json, and CI matches package.json scripts, but most feature claims appear only in README — deducted. Fact/inference separation: README and AGENTS.md mix product claims with internal workflow without distinguishing verified facts from plans — deducted.
- The installer runs curl | sh from a remote URL, and stable releases publish without human confirmation; review the script before running it.
- Agents are routed to Jira, AWS, Vercel and similar tools carrying the user's sign-in, and the sample does not explain credential storage or per-agent isolation.
- AGENTS.md tells agents to commit directly to main and auto-publish, with no unified guardrail for destructive actions; use in a controlled branch or sandbox.
- Publisher identity is unverified, so maintainer and update-channel trust cannot be confirmed.
- Most README feature claims lack code or test corroboration in the sample, so actual behavior cannot be verified statically.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.6.3-nightly.549 Pre-release
Release notes excerpt
Sikemux v0.6.3-nightly.549 Built from [5947124e0](https://github.com/nodelike/sikemux/commit/5947124e0f3e213e53b4048e17c4175719f67179) on main. New since 0.6.3-nightly.542 - stop holding updates for agents in a turn - keep sending updates.blockers to phone 0.2.0 - Windows installers can join a stable release - Windows download builds a tag that carries its version - let agents push clones of outside repositories - ⌘⏎ keeps a call open, ⇧⏎ adds a line …View on GitHub ↗ -
mobile-v0.2.1 Sikemux for Android 0.2.1
FAQ
Does Sikemux itself cost anything?
Does it host models or resell tokens?
Do running agents stop if I close the window?
Can I run it on a server?
Which operating systems are supported?
Is Sikemux open source?
Is Sikemux safe to use?
Is Sikemux still maintained?
Who makes Sikemux?
What is Sikemux's official website?
Thanks — we use this to improve the page.