Not a vibe coder.
It's an engineering tool.
Built for engineers who ship production code — not prompt-and-pray demos. Plan, write, review, and test with full control over every change, using the models you already trust.
Works with the models you already pay for
Anthropic · OpenAI · Google · GitHub Copilot · DeepSeek · Groq · Ollama · xAI · OpenRouter · Mistral
Multi-Agent
Parallel agents. No conflicts.
Plan, Code, Pilot, Review, Advisor — specialized agents running side by side so big tasks finish faster.
Diagrams
Diagrams, rendered in chat.
Ask for a flowchart, sequence, Gantt, or timeline and Kodein draws it inline — 30 Mermaid diagram types rendered natively in pure Dart. No browser, no network, no round-trip.
Code Search
A local semantic index.
Kodein indexes your whole codebase on-device — files, symbols, and vectors — so it finds code by meaning, not just keywords. Embeddings run on your own GPU.
Sign In
Bring your own plan.
Sign in with your existing Claude, ChatGPT, Gemini, or Copilot subscription — or drop in an API key. Kodein tracks your live usage and limits so you always know where you stand.
Usage
Every token, every dollar.
Live context usage, cache savings, and exact cost stay in view — never a surprise bill.
Extend
Connect anything.
Plug in MCP servers, language servers, and a built-in browser. Kodein speaks the protocols your stack already uses.
Memory & Skills
It learns your project.
Kodein remembers decisions across sessions and lets you save reusable skills it can run on demand.
Permissions & Sandbox
Nothing runs without your say-so.
Every command and file write asks first — allow once, always allow, or deny. Work runs sandboxed and scoped to your project, so the agent can move fast without going off the rails.
Multi-Review
Review with a panel of models.
Choose how many reviewers and which model each one runs. They review your diff independently, then a supervisor merges the verdicts — so no single model's blind spot slips through.
Review
Every change, every hunk.
Review what the agent changed and accept or reject per hunk or line — even run a panel of models to review for you.
Multi-Advisor
A second opinion — or three.
Before you commit to an approach, fan the question out to multiple advisors on different models and compare their takes side by side — no edits, just perspective.
Specialized Agents · Multi-Model
One task, many minds.
Purpose-built agents with their own tools and prompts — and for the calls that matter, a whole panel of models instead of a single opinion.
Any Model
Use the plan you already pay for.
Sign in with your subscription — or drop in an API key. Same agent, every provider, no markup.
Any OpenAI-compatible endpoint works too — point Kodein at your own inference server.
Platforms
Native everywhere.
No Electron. No webview.
macOS
Native NSWindow tab groups. Single-instance, GPU-accelerated.
Linux
Native GTK. Terminal support, filesystem access, Ollama local inference.
Windows
Win32 with Mica/Acrylic. CUDA + Vulkan for local GPU inference.
VS Code
Full Flutter UI inside VS Code via gRPC-Web. Same agent, same tools.