Oblivion v2.9.0 — First public release 🚀
Oblivion is now a real distributable Python package. Install with pip, talk to it with voice, use it from Claude Desktop via MCP.
✨ Highlights
- 🎙️ Voice in & out — press Ctrl+T, talk to Meera, hear her reply (Whisper + Edge TTS / ElevenLabs)
- 🤖 9 LLM backends — Gemini, Groq, Claude, GPT-4o, DeepSeek, Ollama, Cerebras
- ⚡ Auto-fallback — silent model swap when one rate-limits
- 🧠 Hybrid code search — symbol lookup + full-text + semantic embeddings
- 🔌 MCP server — Claude Desktop can use Oblivion's 10 read-only tools
- 📚 Knowledge packs — React, Django, FastAPI, Tailwind, Docker, security, more
- 📜 Session history — Ctrl+G to resume any past conversation
- 🌃 Cyberpunk TUI — Textual-based with animated boot sequence
📦 Install
```bash
pip install oblivion-agent
```
With voice support:
```bash
pip install "oblivion-agent[voice]"
```
🚀 First Run
```bash
oblivion
```
Setup wizard walks you through picking an LLM provider (Gemini recommended — free + 1500 req/day).
🔌 Claude Desktop Integration
Add to claude_desktop_config.json:
```json
{
"mcpServers": {
"oblivion": {
"command": "oblivion",
"args": ["mcp"],
"env": { "WORKSPACE_DIR": "/your/project" }
}
}
}
```
📋 What's New Since v2.0
- Pip-installable package structure (~/.oblivion config dir)
- Auto-fallback chain for LLMs
- MCP server with 10 read-only code tools
- Session history picker (Ctrl+G)
- Subcommand CLI (oblivion mcp, oblivion init)
- Dynamic /model autocomplete (all 9 models)
- Voice transcription bias for proper name recognition
Built with ❤️ by R. Rohit.