Intent Agent Workbench
Coordinate coding agents, repositories, and development work from one desktop workspace.
- Source repo
- intent-hq/intent
- Stars
- ★ 300
- Last updated
- today
- License
- Apache-2.0
- Primary language
- JavaScript
- FA score
- 60/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Developers who want to manage multiple coding-agent workspaces from one desktop environment.
- Not a fit if
- Teams that need the desktop app source to be publicly auditable
- Developers who need to submit external pull requests now
- Users who need a public iOS app repository
- Source review
- 60/100 · Some gaps
What does this agent do, and when should you use it?
Intent is a desktop application for coordinating coding agents at scale, bringing workspaces, notes, tasks, agents, Git, and terminals together. Its Electron and SvelteKit desktop client uses intentd, a Rust daemon that owns state and business logic and persists data in SQLite. Local clients communicate over JSON-RPC 2.0 on a Unix-domain socket; LAN clients such as the iOS companion app can use WSS/TLS. The packaged desktop app bundles and manages intentd, while the daemon can also be deployed independently; this repository connects the intentd, cloudlands-fe, and iOS submodules, with component code living in their respective repositories. The project describes itself as local-first and says it ships no telemetry, analytics, or crash reporting; the iOS companion app is early in development and its repository is private.
The Intent desktop app uses intentd to manage workspaces, notes, tasks, coding agents, Git, and terminals. Running locally, intentd persists state to SQLite and serves local clients through JSON-RPC 2.0 over a Unix-domain socket; when WSS is enabled, it can also serve LAN clients over TLS. The desktop app can spawn and supervise its bundled daemon or be configured to connect to an already-running one. For standalone deployments, the intentd sitter downloads and launches the real daemon, forwards CLI arguments, and respawns it after a crash. Users can also configure GitHub, Linear, Sentry, MCP servers, or coding-agent providers; those connections call the relevant service or run an external agent program.
- Developers who want to manage multiple coding-agent workspaces from one desktop environment.
- Engineering teams that want tasks, notes, Git, and terminals alongside agent workflows.
- Developers who want to run the daemon locally and connect a desktop client to it.
- Users who need an iOS companion client to connect to intentd over a LAN.
- Developers who want to run intentd independently on a remote host and connect a desktop app to it.
How do you install or deploy this agent?
The recommended route is to download the packaged desktop app from intentapp.dev; it manages its bundled intentd, so a separate daemon installation is unnecessary. To run the daemon independently, install the sitter with the macOS/Linux script below and then start the service. For a source build, clone the repository with its public submodules; the private iOS submodule is skipped automatically.
How do you use this agent?
For a first run, open the packaged app and let it start and supervise its bundled intentd. If you installed intentd separately, start the service with the command below. With Homebrew, you can register it as a background service. The desktop app can also be configured to connect to an already-running daemon. Network access is disabled by default; LAN WSS must be enabled and configured, and clients can discover the actual listening port with intentd pair or the pairing payload.
What are this agent's strengths and limitations?
- The desktop app can bundle and supervise the Rust daemon, avoiding a separate intentd installation.
- State and business logic live in a local Rust daemon and persist to SQLite.
- Local clients use a Unix-domain socket, while LAN clients can use WSS/TLS.
- The README explicitly says the product ships no telemetry, analytics, or crash reporting.
- Core code is split across the intentd, cloudlands-fe, and iOS submodule repositories; this repository mainly carries documentation, tooling, and CI.
- The cloudlands-fe and iOS repositories are private submodules, and external contributions are currently deferred.
- The iOS companion app is still in early development.
- Coding agents are external programs whose network behavior depends on their providers; provider sign-ins and integrations require user configuration.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Intent Agent Workbench This agent | 60 · Some gaps | Desktop app | ★ 300 | today | JavaScript | — |
| Foremerge | 77 · Good | CLIFree | ★ 538 | 5d ago | Rust | Codex · Claude Code |
| Keep the Why | 75 · Good | Agent plugin / skillFree | ★ 207 | today | Python | Codex · Claude Code |
| Superset | 52 · Major gaps | Desktop appFreemium | ★ 15k | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes local-first operation, disabled-by-default network access, update checks, user-initiated provider sign-ins, and integrations that require user configuration. This supports understanding data flows and external effects. Deductions: coordinating coding agents, terminals, Git, and user-supplied tokens can involve high-impact actions, but the supplied files do not describe per-action permission controls or confirmation gates; credential protection, rollback, and dependency-security practices are also only partially covered. The README, SECURITY.md, and workflow identify components, reporting paths, and some automation, while publisher identity remains unknown.
The README gives coherent guidance for local sockets, WSS port assignment, installation, and development commands; it describes startup failure on port binding and guidance when the daemon is not yet installed. Deductions: these are documented behaviors without runtime evidence in the supplied static materials; the project depends on submodules and external release channels, and the private iOS submodule is skipped.
The README addresses desktop users, standalone daemon users, and developers, with local, remote-host, and multiple development modes; it also describes boundaries around external coding agents and configured integrations. Deductions: the supplied materials do not establish precise trigger conditions or detailed boundaries for agent capabilities; the iOS app is identified as early-stage.
The README organizes architecture, installation, networking and privacy, building, and community information clearly. Installation steps cover several platforms and explain port behavior; LICENSE contains the Apache-2.0 terms. SECURITY.md describes supported versions, project scale, and vulnerability-report expectations. Deductions: the supplied files contain no formal release history, naming stability is mostly inferred from documentation, and maintenance has a contact path but no clearly identified individual or team.
The README presents a concrete use case: coordinating workspaces, tasks, agents, Git, and terminals through a desktop app with centralized daemon state, and provides actionable install and development commands. Deductions: the supplied materials contain no usable output examples, demo, or run results to substantiate the value relative to its costs.
The README and SECURITY.md make specific product, installation, network, and security claims; the workflow files show intended submodule-bump and installer-check processes. The documentation distinguishes some facts from project-stage limitations. Deductions: cross-source corroboration is limited, key component code is in submodules whose contents were not supplied, and documentation cannot replace execution evidence.
- This low-confidence assessment uses only the supplied static files; no installation, agent operation, or tests were run.
- The README says coding agents are external programs whose network behavior depends on their providers; connecting services uses credentials supplied by the user.
- Core code resides in submodules, and the iOS submodule is private; their implementation details could not be reviewed here.