SuperCoder v0.1.0 — local-first coding agent

✨ SuperCoder v0.1.0 ✨

A local-first, open-source coding agent for your desktop

SuperCoder has been reimagined from the ground up — a local-first, open-source coding agent for your desktop.

💡 What it is

A desktop app built on a pure-Rust agent harness. Your code stays on your machine and only ever leaves to the LLM provider you configure — no middleman service, no lock-in.

• 🔒 Local-first & BYO-key — native OpenAI and Anthropic support (no translation proxy); bring your own endpoint and key.
• 🧭 Ask / Plan / Coding modes, subagents, skills, and tool approval.
• ↩️ In-place edits with a safety net — every turn is checkpointed; rewind restores exactly the files the agent touched.
• 🕸️ Graph-aware code retrieval (optional Context Engine) — indexes your repo into vector + call-graph + lexical search so the agent navigates large codebases by structure, not just text similarity.

🔎 Graph-aware code retrieval

Most agents retrieve by embedding similarity alone. The Context Engine indexes your repo with tree-sitter into three complementary stores — Qdrant (vectors), FalkorDB (call/symbol graph), and BM25 (lexical) — so the agent can locate code structurally (codebase_search / codebase_graph), not just by what reads similar. Incremental Merkle streaming sync keeps the index current as you work. The app manages the whole Docker stack for you — pulls the versioned images, brings it up, health-checks, and tears it down on quit.

🚀 Two ways to run

• Zero-backend (default): add an LLM key and go.
• Context Engine (opt-in): enable it in Settings to run a local Docker stack the app manages for you.

📦 Install

Grab the build for your OS below:

• macOS — .dmg
• Windows — .exe installer
• Linux — .AppImage / .deb

---

See the README to get started, and Discussions for questions and ideas.

_{First release of the new SuperCoder — early, and moving fast. Issues and contributions welcome. 💛}