The open source coding agent for your terminal.
Explore a codebase, plan a change, and work across files with explicit approval for writes and commands. Start read-only in Plan mode, then switch to Agent when you are ready to build.
Requires the .NET SDK 10.0 or later.
dotnet tool install --global CodaiRun from your project directory:
codai # First run: choose theme, provider, and model; later runs open chat
codai auth # Change the connection or selected modelNew setups save user-wide defaults in ~/.codai/codai-config.json when no project
configuration exists. See first-run setup.
| Capability | How it helps |
|---|---|
| Plan, then act | Read and search without changing files or running commands; switch to Agent for implementation. |
| Project-wide edits | Read, search, and edit real files with your project's guidance in context. |
| On-demand context | Open sessions without a full-project indexing pass; discover files through tools as needed. |
| Controlled execution | Review writes and commands under your approval policy. Shell commands require shell access. |
| Focused terminal UI | Scroll the transcript with input pinned; expand action steps, command output, and file diffs. |
| Your choice of models | Connect hosted providers or local servers, including Ollama. |
| Reusable workflows | Apply project guidance, rules, skills, commands, and memory documents. |
| Long-running work | Compact context and resume a previous session with codai code --continue. |
| Extensible tools | Add MCP servers, .NET plugins, and lifecycle hooks under the same approval policy. |
Connect through codai auth, or select a provider with --provider and a model with
--model.
| Provider | --provider |
Hosting | Default model |
|---|---|---|---|
| OpenAI | openai |
Cloud | gpt-5.4-mini |
| Azure OpenAI | azure-openai |
Cloud | Your deployment ID |
| Anthropic | anthropic |
Cloud | claude-sonnet-4-6 |
| Google Gemini | gemini |
Cloud | gemini-3.8-flash |
| Ollama | ollama |
Local | qwen3-coder |
| DeepSeek | deepseek |
Cloud | deepseek-flash |
| OpenRouter | openrouter |
Cloud | openai/gpt-5.4-mini |
| Qwen (DashScope) | qwen |
Cloud | qwen3.7-plus |
| Mistral | mistral |
Cloud | mistral-medium-3-5 |
| Grok (xAI) | grok |
Cloud | grok-4.7 |
Hosted providers require API credentials. Local Ollama does not require an API key.
OpenAI also supports Continue with ChatGPT for eligible accounts through codai auth;
account-plan access and API credentials are separate connections.
Model availability and coding-tool support depend on the selected provider and model.
Setup fetches live model catalogs, with bundled suggestions when discovery is unavailable.
See model selection and compatibility.
Bundled model choices include provider-specific USD estimates and prompt-length tiers;
see pricing coverage and limitations.
Documentation uses Nextra, with Codai's default blue, light/dark themes, local
full-text search, and guided setup. GitHub Actions builds and publishes the static site
to GitHub Pages when documentation changes reach main; no hosted docs account is
required. See documentation development for local preview.
| Guide | Covers |
|---|---|
| Overview | Introduction to Codai |
| Quickstart | Installation, model connection, and your first session |
| Agent mode | Modes, approval policy, terminal controls, and session recovery |
| Skills | Project and global skills |
| Plugins | Custom tools and lifecycle hooks |
| MCP integrations | External tool servers |
| ACP compatibility | Supported editor integration and current limits |
| Roadmap | Planned work and implementation status |
Bug reports, feature requests and pull requests are welcome. Start with CONTRIBUTION.md; report security issues through