🌌 Stepbit: The High-Performance LLM Command Center

Stepbit is a premium, local-first LLM orchestration platform built with a Rust backbone and DuckDB intelligence. It’s designed for developers who demand speed, security, and a beautiful interface for their AI workflows.

Stepbit isn't just a server; it's a sleek, glassmorphic experience inspired by the Monokai-Gruvbox palette.

• Live CSS & HTML Rendering: Inject active styles and raw HTML directly into chat for interactive demos.
• Rich Markdown Rendering: Full GFM support with optimized syntax highlighting.
• High-Fidelity Charts: Interactive Line and Bar charts with sub-micro granularity. Learn more
• Live SVGs: Dynamic graphic rendering directly in the chat.
• Micro-animations: A fluid, responsive interface with real-time Thinking & Searching status.
• Process Cancellation: Instantly stop any long-running generation or search.

• Rust Core: Blazing fast, memory-safe execution using Actix-web.
• DuckDB Storage: Analytical conversation memory in a single file (chat.db). Learn more
• Pluggable Intelligence: Seamlessly switch between OpenAI, Anthropic, Ollama, GitHub Copilot, and the optional stepbit-core Cognitive Engine.
• Chained Request Security: Rolling handshake protection for internal stepbit-core communication, preventing replay attacks.
• DuckDB Snapshotting: Non-blocking analytical access to live data. Learn more
• Reasoning Graph Engine: A powerful DAG-based architecture with Server-Sent Events (SSE) for real-time node lifecycle tracking.
• Cognitive Pipelines: Structured, deterministic workflows with multi-stage reasoning, MCP integration, and incremental result streaming.
• MCP Tool Registry: Integrated Model Context Protocol support for tool discovery and schema management.
• Real-Time Dashboard: High-fidelity telemetry for tokens, messages, and storage.
• Node 24 & React 19: Modern frontend stack for maximum response times and security.

Cognitive Pipelines are the core of Stepbit's automated reasoning. Here is how to get started in 3 steps:

1. Enable stepbit-core: Ensure stepbit-core is running (optional but recommended for advanced pipelines).
2. Import a Pipeline: Go to the Pipelines tab and click New Pipeline. Paste a JSON definition.
3. Execute: Ask a question against the pipeline and watch the Reasoning Trace unfold in real-time.

git clone https://github.com/jacovinus/stepbit.git
cd stepbit
cargo build --release

Edit config.yaml to unleash your preferred models:

llm:
  provider: "ollama"  # or "openai", "anthropic", "copilot"
  model: "llama3.2"

cargo run -- serve

• AI Command Center: http://localhost:5173 (Primary Vite/React interface)

Switch providers and models on the fly without restarting the server. The dynamic sidebar and chat header allow for instant context switching.

Build and execute structured AI workflows with deterministic stages. Monitor every step of the reasoning process through the live Trace Viewer.

• Multi-Stage Orchestration: Combine LLM generation, MCP tool calls, and automated verification.
• Deep Visibility: See exactly how the AI arrived at an answer with granular trace logs.

• Database Purge: Clear all data instantly via cargo run -- database purge or API.
• Export/Import: Move your AI brain between environments via .txt files. Learn more
• Skills Library: Reusable prompt snippets and persona templates. Learn more

• Features Guide: Deep-dive into capabilities and use cases.
• Changelog: Latest updates and version history.
• Project Journey: Roadmap & vision board for the future.
• API Guide: Detailed REST & WebSocket endpoint documentation.

Built with 🦀 and 🎨 for those who care about the details.