goose vs trueforge
General-purpose local AI agent in Rust — native desktop app, full CLI and embeddable API — over 15+ providers and 70+ MCP extensions. Block's goose, now an Agentic AI Foundation project. — versus — TrueFoundry's open agent harness: the runtime loop — model calls, MCP tools, SKILL.md packs, sandboxing, approvals, compaction — behind a chat UI, HTTP API, TypeScript SDK and embeddable UI.
Both are complete harnesses — model loop, MCP tools, SKILL.md packs, sandboxing, approvals, compaction. Trueforge is server-first with an HTTP API, TS SDK and embeddable UI for putting an agent inside your product; goose is a native app and CLI for putting an agent on your machine.
| goose | trueforge | |
|---|---|---|
| Stars | 54k | 4.7k |
| Forks | 6.1k | 305 |
| Language | Rust | TypeScript |
| License | Apache-2.0 | MIT |
| Last activity | today | yesterday |
| Topics | agents, coding | agents, orchestration |
| Curated connections | 5 | 5 |
goose — the curator's take
The default answer when someone wants a real agent on their own machine and does not want to assemble one. Rust, three surfaces (desktop app, CLI, API), 15+ providers including Ollama and your existing Claude/ChatGPT/Gemini subscriptions over ACP, 70+ MCP extensions, and custom distributions if you want to ship your own branded build. Now governed under the Linux Foundation's AAIF rather than one vendor, which matters if you are betting a product on it. It is deliberately general — code, research, writing, data — so it has no opinion about your SDLC: no verification gates, no worktree isolation, no PR pipeline. If you want an agent that must prove its work, or a spec-to-PR factory, layer that on or pick a purpose-built harness. Note the repo moved from block/goose; old links and forks still point at the old org.
trueforge — the curator's take
Take it when the agent logic is the easy part and you don't want to hand-build streaming, session persistence, tool auth, sandboxing and an approval UI: YAML catalogs configure models, MCP servers, skills and a sandbox once, then agents pick from what you connected. Context engineering is first-class — subagents, deferred tool loading, Code Mode, large-result offloading, compaction — and it scales from `npx` on SQLite to Postgres+Redis via Docker Compose or Helm. Caveats: local mode has no login and is explicitly not for anything internet-facing, sandboxing means Daytona today, and it's a harness rather than a graph framework — if you need bespoke control flow per step, a lower-level runtime fits better.