← all repositories
codestoryai/sidecar

Aide's AI brain runs locally, in Rust, with a pagerank map of your code

Sidecar is the backend that turns the Aide editor into a self-hosted coding assistant—no cloud required, no subscription implied.

sidecar
Velocity · 7d
+0.6
★ / day
Trend
steady
star history

What it does

Sidecar is the Rust backend that powers Aide, an AI-assisted code editor. It handles prompt construction, LLM communication, and tool orchestration—all running locally on your machine. Think of it as the server process that lets an editor talk to language models without phoning home to a vendor’s API for the heavy lifting.

The interesting bit

The project borrows a clever trick from Aider: it builds a repository map using PageRank on code symbols, so the AI knows which files and functions actually matter in your codebase. The symbol/ module also lets the system reason at any granularity—from whole files down to individual methods—rather than treating code as flat text.

Key highlights

  • Self-hosted by design: build the binary, run the webserver, point Aide at it
  • tool_box.rs centralizes language-aware tooling for the LLM to call
  • repomap uses PageRank on code symbols to prioritize context (ported from Aider’s Python implementation)
  • AGPL v3 licensed—open source, copyleft
  • Requires Rust 1.79; builds to a single webserver binary

Caveats

  • Inline completion prompts in llm_prompts/ are explicitly unmaintained; the team says to reach out on Discord if you want to revive them
  • The README is sparse on architecture details—how the LLM is queried, which models are supported, and how the tool loop works are left unstated

Verdict

Worth a look if you’re building a local-first AI editor or want to study how to structure a Rust backend for LLM tool use. Skip it if you need a drop-in Copilot replacement without also adopting the Aide editor; this is tightly coupled to its frontend.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.