A full-featured PyQt6 code editor where AI doesn't just suggest — it reads, writes, runs and verifies code on your behalf through a structured command bridge.
AI doesn't just chat — it operates the editor through a structured JSON protocol
The AI reads a file once with get_file_content, holds the content in context, then issues multiple anchor-based edits without re-reading between each change. No wasted round-trips.
Edits reference exact text to find, not line numbers. Line numbers shift as code is added or removed — anchors don't. Multiple replace_text calls can be batched in a single response, all applied atomically.
For new modules or complete rewrites the AI sends the entire file content in one command. The system automatically creates a .bak backup before overwriting and opens the result in a new editor tab.
The AI can run shell commands and receive the JSON result before continuing. It waits for each result, handles exit codes, and distinguishes diagnostic, configuration, destructive and session-ending commands automatically.
All content returned from files, terminals and web pages is wrapped in [UNTRUSTED_CONTENT] markers. The AI is instructed to treat everything inside as data — never instructions — preventing prompt-injection attacks from malicious file content.
Open multiple files simultaneously. Each tab tracks its own undo history, syntax highlighting and AI command context.
One click to compare the current editor state against the last saved version. Colour-coded unified diff with add/remove highlighting. Exit diff mode to return to full editing.
Responses stream token-by-token into the chat panel. IDE commands are extracted and executed in a background thread so the UI stays responsive throughout.
System prompts are plain-text module files in ~/.chameleon/prompts/. Stack any combination: Tech Repair + Academic + Chat. Live token count and cost estimate per provider.
Every AI command is logged to history.db with session ID, timestamp and result. Full audit trail for every file change the AI made.
Project tree panel for fast navigation. The AI uses list_project_files and list_open_files to discover real paths — it never invents file locations.
Local Whisper (whisper.cpp) transcribes your voice into the chat input. Hands-free coding — dictate questions, commands or whole prompts without touching the keyboard.
AI responses are read aloud using Piper TTS — fast, local, natural-sounding voices. Ideal for accessibility or working with your hands occupied. Platform paths resolved automatically on Linux, Windows and macOS.
| Provider | Type | Cost (approx.) | Notes |
|---|---|---|---|
| DeepSeek | Cloud API | $0.00014 / 1K tokens | Default; excellent code quality |
| OpenAI (GPT-4o) | Cloud API | $0.0050 / 1K tokens | Best general reasoning |
| Anthropic (Claude) | Cloud API | $0.0030 / 1K tokens | Strong at long-context editing |
| Claude Code (Anthropic) | Terminal / pty | API usage | Runs as a pty process inside the IDE terminal — full agentic coding in a live shell |
| Groq | Cloud API | $0.00005 / 1K tokens | Ultra-fast inference |
| Mistral | Cloud API | $0.00025 / 1K tokens | Good European-hosted option |
| Moonshot / Kimi | Cloud API | $0.00012 / 1K tokens | Large context window |
| Ollama | Local | Free | Run any GGUF model locally |
| LM Studio | Local | Free | GUI-friendly local inference |
Coming soon — IDE in action
Contact us for a live demo or to discuss a licence for your team.
Request a demo ← Back to overview