A terminal-based AI coding assistant with multi-provider support, screenshot analysis, and a full-featured TUI.
Run any AI model from your terminal — with tools for file editing, bash, search, fetch, vision analysis, LSP, MCP, and more. Built for developers who want a fast, local AI coding experience.
Quick Start · Features · CLI Usage · Configuration · Dashboard · Contributing
git clone https://github.com/chrisbeckett/talon.git
cd talon
# Set at least one API key
export ANTHROPIC_API_KEY=sk-ant-...
# or
export OPENAI_API_KEY=sk-...
# or
export OPENROUTER_API_KEY=sk-or-v1-...
# Build and run
go build -o talon .
./taloncd docker
docker compose up -d| Feature | Description |
|---|---|
| Terminal UI | Full-featured Bubble Tea TUI with sidebar, chat view, file tree, and session management |
| Multi-provider | OpenAI, Anthropic, OpenRouter, Google Gemini, Groq, DeepSeek, Ollama, and more |
| File editing tools | View, write, edit, multi-edit files with LSP integration |
| Bash tool | Run shell commands with permission management and sandboxing |
| Search tools | Glob, grep, ripgrep, web search, and sourcegraph code search |
| Fetch tool | Fetch web pages and convert to markdown |
| Screenshot capture | Capture your screen and analyze it with AI |
| Vision analysis | Local MiniCPM-V integration for image/document analysis via Ollama |
| MCP server support | Connect Model Context Protocol servers for expanded tool capabilities |
| LSP integration | Language Server Protocol integration for diagnostics and references |
| Skills system | Extensible skill-based tool loading |
| Sub-agents | Delegate tasks to specialized sub-agents |
| Session management | Persistent conversation history with automatic summarization |
| Context-aware | Automatic context management with token tracking and memory tree compression |
| Client/Server mode | Run as a daemon on a Unix socket with HTTP/SSE streaming |
| Hooks system | Gate, approve, or rewrite tool calls before execution |
| CVE database | Built-in CVE vulnerability search and tracking |
| Token optimization | Response caching, token optimization, and context compression |
| Dashboard | Web dashboard with live status, task stats, and configuration |
| Docker support | Containerized deployment with docker-compose |
# Start the TUI
./talon
# Direct prompt
echo "list all go files in this project" | ./talonOnce in the TUI:
Ctrl+P— Open command paletteCtrl+D— Toggle details panelCtrl+N— New sessionTab— Switch focus between editor and chat
Configuration is stored in ~/.talon/talon.json. The first run will guide you through provider setup.
For screenshot analysis and image understanding with models that don't support vision natively:
# Install MiniCPM-V via Ollama
ollama run minicpm-vConfigure in talon.json:
{
"tools": {
"vision": {
"endpoint": "http://localhost:11434/v1/chat/completions",
"model": "minicpm-v"
}
}
}No configuration needed for natively vision-capable models (Claude, GPT-4o, Gemini).
| Dashboard | Tasks / Kanban |
 |
 |
| Memory Vault | Model Selection |
 |
 |
| Settings / Configuration | Floating Bottom Bar |
 |
 |
| Provider | API Key |
|---|---|
| Anthropic | ANTHROPIC_API_KEY |
| OpenAI | OPENAI_API_KEY |
| OpenRouter | OPENROUTER_API_KEY |
| Google Gemini | GEMINI_API_KEY |
| Groq | GROQ_API_KEY |
| DeepSeek | DEEPSEEK_API_KEY |
| Mistral / Codestral | MISTRAL_API_KEY / CODESTRAL_API_KEY |
| Azure OpenAI | AZURE_OPENAI_API_KEY (with endpoint) |
| AWS Bedrock | AWS credentials |
| Ollama (local) | None (runs locally) |
| llama.cpp / LM Studio | None (runs locally) |
| OpenCode | OPENCODE_API_KEY |
| Fireworks AI | FIREWORKS_API_KEY |
| Cerebras | CEREBRAS_API_KEY |
| NVIDIA NIM | NVIDIA_NIM_API_KEY |
| Kimi / Moonshot | KIMI_API_KEY |
| Z.AI | ZAI_API_KEY |
| Wafer | WAFER_API_KEY |
Set API keys in ~/.talon/.env:
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
OPENROUTER_API_KEY=sk-or-v1-...talon/
├── internal/
│ ├── agent/ # Agent orchestration, tools, prompts
│ │ ├── tools/ # Tool implementations (bash, edit, view, etc.)
│ │ ├── hyper/ # Provider capability definitions
│ │ └── templates/ # System prompt templates
│ ├── app/ # Application wiring and lifecycle
│ ├── backend/ # Transport-agnostic workspace/agent management
│ ├── client/ # Client for connecting to server
│ ├── cmd/ # CLI entry points (Cobra)
│ ├── config/ # Configuration management
│ ├── db/ # SQLite (sqlc-generated queries)
│ ├── hooks/ # PreToolUse hook runner
│ ├── lsp/ # LSP client manager
│ ├── message/ # Message CRUD (debounced writes)
│ ├── permissions/ # Permission approval dialogs
│ ├── pubsub/ # Typed pub/sub event broker
│ ├── server/ # HTTP/SSE server over Unix socket
│ ├── session/ # Session CRUD service
│ ├── shell/ # Cross-platform shell (mvdan/sh)
│ ├── skills/ # Agent skills discovery
│ ├── ui/ # Terminal UI (Bubble Tea)
│ │ ├── chat/ # Chat message rendering
│ │ ├── dialog/ # Dialog components
│ │ ├── model/ # UI models (header, sidebar, etc.)
│ │ └── styles/ # Theming and styles
│ ├── version/ # Version information
│ └── workspace/ # Workspace abstraction (local vs client/server)
├── deps/ # Forked Charm dependencies
├── docker/ # Docker configuration
├── dashboard/ # Next.js web dashboard
└── dashboard/public/ # Assets, favicon, screenshots
Talon is in active development. Contributions are welcome.
go build ./... # Verify compilation
go test ./internal/... # Run tests- 🧪 Test providers — Configure different providers and report issues
- 🐛 Bug reports — Include model, error message, and reproduction steps
- 🔧 New tools — The tool system is extensible; add your own tools
- 🎨 UI improvements — The TUI is built with Bubble Tea; contributions welcome
- 📖 Documentation — Improve docs, add examples
MIT License. See LICENSE for details.