Show HN: Aide, an open-source AI native IDE

Original Article ↗ Hacker News Discussion ↗

An open-source, AI-native IDE called Aide—built as a fork of VS Code with a separate “Sidecar” AI engine—aims to act as a true pair programmer by deeply integrating code editing, multi-file refactoring, and local or user-controlled LLMs. Commenters weigh its advantages against tools like Cursor, Copilot, Zed, Cline, and Aider, focusing on trade-offs between IDE forks vs plugins, UX ergonomics, performance, and pricing models. Early users report rough edges around onboarding, latency, and naming confusion, but see promise in its open licensing, privacy controls, and potential for broader model and editor integrations.

Project basics & naming

  • Aide is a VS Code fork with an AI “sidecar” process; both are open source on GitHub (IDE + Rust sidecar).
  • Sidecar is fully open source; name overlaps with Apple’s Sidecar and Kubernetes sidecars, but many consider such duplication inevitable.
  • “Aide” conflicts with an Android IDE and the long‑standing AIDE intrusion detection project. Some argue for renaming; others say the use cases are distinct and the AI+IDE pun is too good to drop.

Differentiation vs Cursor, Copilot, Zed, etc.

  • Team claims advantages over other AI IDEs:
    • Deep editor integration (not just an extension).
    • Everything can run locally; users control LLM providers and keys.
    • Strong rollbacks that preserve the editor’s native undo/redo.
    • Open‑source licensing and data ownership.
  • Critics note Cursor already has deep integration, rollbacks, and workflows; they press for a clearer “killer feature” beyond Sidecar.
  • Some suggest VS Code extensions or CLI tools are more sustainable than maintaining a full fork; others report that AI‑native forks like Cursor feel significantly better than plugins.

Architecture, models, and context

  • Sidecar is a Rust “AI brain” using tree‑sitter and LSP for symbol understanding and context building.
  • The system issues smart go‑to‑definition/reference queries and applies heuristics to limit LLM calls and latency.
  • Adding new LLM backends (e.g., Claude via AWS Bedrock, Qwen 2.5) is described as straightforward; contributors are invited to add clients.

UX, stability, and platform support

  • Early users report:
    • Broken first‑run installer UI, confusion over “Trusted mode.”
    • High latency and global queuing when using the project’s shared API keys.
    • Confusing free‑request limits and a buggy login flow.
  • Maintainer attributes latency to launch traffic and rate limits, suggests using personal API keys, and acknowledges auth/UX issues.
  • Download links briefly broke due to GitHub rate limiting; fixed later. Linux .deb and scripts are available; AppImage confirmed working.
  • Some find the @/pinning context UX non‑intuitive and want automatic inclusion of the current file and better multi‑file context ergonomics.

Privacy and telemetry

  • Users ask about code/secret exposure. Maintainer says no telemetry is sent by default; a setting can explicitly disable telemetry, which short‑circuits the analytics client.

Broader AI coding discussion

  • Thread branches into wider tool comparisons: Cursor, Copilot, Zed AI, aider, Cline, and browser‑based ChatGPT/Claude.
  • Themes: AI excels at autocomplete and structured, repetitive tasks; multi‑file edits and reliability remain challenging; cost models, UX preference (in‑editor vs browser), and language support heavily influence tool choice.