Everything a serious AI workspace needs.
Provider‑neutral by design, local‑first by default, and safe by construction.
Any model, per role
Assign chat, edit, agent, completion, and embedding roles to different models — local, LAN, or cloud.
Ollama support
On‑device inference as a first‑class path. Point at any Ollama endpoint on your machine or network.
Agent mode
Structured tool calls with universal permission review, once/session/workspace scopes, revocation, and an audit trail.
Indexing & retrieval
Lexical retrieval by default; hybrid with an embedder. A transparent context manifest for every request.
Inline edit
Structured diffs, per‑hunk acceptance, atomic apply with baseline checks, and operation‑level undo.
Completion
Native ghost‑text with Tab accept, Escape dismiss, cooldown, and repetition prevention.
Integrated terminal
A real PTY for you, kept strictly separate from the agent's bounded, non‑interactive command executor.
Secure transport
All model traffic runs through the trusted Rust host. Keys live in the OS keychain; the webview never sees them.
Local‑only mode
A policy gate blocks cloud and unapproved LAN requests — enforced in code, with no silent fallback.
VS Code + desktop
Shared workspace language across a VS Code extension (no account) and a native Tauri desktop app.
MCP servers
Add third‑party tools via Model Context Protocol — disabled by default, per‑tool allowlists, outputs treated as untrusted.
Audit & permissions
Every action is validated, risk‑classified host‑side, approved, and recorded as redacted metadata.