@kolaborate/praxis
v0.2.2
Published
An offline-first AI coding agent with browser automation
Readme
Praxis 🛰️
Praxis is a high-performance, offline-first AI coding agent built in Rust. It utilizes a dual-model orchestration approach powered by Ollama to perform complex tasks, manage browser automation, and execute multi-step reasoning loops.
✨ Key Features
- 🧠 ReAct Reasoning Loop: Iterative Thought → Action → Observation pattern for autonomous problem solving.
- 🎭 Dual-Model Orchestration: Specialized "Orchestrator" for tool selection and "Executor" for code generation.
- 🌐 Browser Automation: Seamless integration with
agent-browserfor web navigation, filling forms, and scraping. - ⚡ Parallel Execution: Concurrent tool execution using Tokio
JoinSetfor maximum efficiency. - 🤝 Sub-Agent Architecture: Capability to spawn specialized sub-agents for delegation and parallel task processing.
- 🛠️ Extensible Toolset: Built-in support for coding (write, explain, debug), context analysis, and web tools.
- 🚀 High Performance: Built with Rust for safety, speed, and minimal resource footprint.
📋 Prerequisites
- Ollama: Install from ollama.ai.
- Required models:
qwen3-vl:8b(orchestrator),qwen3:8b(executor). - Setup:
ollama serve ollama pull qwen3-vl:8b ollama pull qwen3:8b
- Required models:
- agent-browser (Optional): For web automation features.
npm install -g agent-browser agent-browser install
🚀 Installation
npm (Recommended)
npm install -g @kolaborate/praxisHomebrew (macOS)
brew tap kolaborate-platforms/tap
brew install praxisDirect Download (Shell Script)
macOS / Linux:
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/Kolaborate-Platforms/praxis/releases/latest/download/praxis-installer.sh | shWindows (PowerShell):
irm https://github.com/Kolaborate-Platforms/praxis/releases/latest/download/praxis-installer.ps1 | iexFrom Source
cargo install --path .⚙️ Configuration
Praxis looks for configuration in ~/.config/praxis/config.toml. It also respects environment variables like OLLAMA_HOST and OLLAMA_PORT.
Example config.toml
[ollama]
host = "localhost"
port = 11434
[models]
orchestrator = "qwen3-vl:8b"
executor = "qwen3:8b"
[agent]
max_history = 1000
max_turns = 10
debug = false
[browser]
enabled = true
session_name = "praxis"
[streaming]
enabled = true🚀 Usage
Interactive REPL:
praxisSingle Prompt:
praxis -p "Research the latest Rust 1.84 features and summarize them."Debug Mode:
PRAXIS_DEBUG=true ./target/release/praxis🧪 Testing & Benchmarking
Praxis includes a benchmark harness to compare different models.
Run All Tests:
cargo testRun Model Benchmarks:
cargo test --test model_benchmark -- --ignored🗺️ Project Structure
src/agent: Core agent logic, conversation management, and ReAct loop.src/llm: Client implementations for Ollama.src/tools: Tool registry and individual tool implementations (Coding, Browser, etc.).src/core: Shared types, error handling, and configuration logic.src/cli: CLI interface and REPL implementation.
📄 License
MIT License - See LICENSE for details. (Place holder)
Possible Future Integrations
- https://aspectron.org/en/projects/workflow-rs.html
- https://github.com/workflow-rs/workflow-rs
