Codai CLI
A terminal agent explores projects, prepares plans and makes edits or runs commands under a chosen approval policy.
- 373 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Agent mode edits files and runs permitted commands; cloud models receive submitted project context.
Install
Manual install
dotnet tool install --global codaiFirst Installation command; installs the global .NET tool.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Agent mode edits files and runs permitted commands; cloud models receive submitted project context.
This is third-party code. Review the repository files before installing.
What it does
Codai works on code from the terminal. In Plan mode it reads and searches without editing or running commands; the user switches to Agent for implementation. Actions, command output and file diffs can be expanded in the conversation. Context is discovered on demand without mandatory full-project indexing. Users connect a cloud model or local server and retain project guidance, skills and memory across sessions.
Who it is for. Developers using .NET or other project stacks who want read-only investigation before allowing implementation.
Good fit when
- Before a fix, you investigate an unfamiliar project.
- With a plan approved, you want the agent to make a focused patch.
- After a break, you need to resume a long task from an earlier session.
- For one workflow, you compare a cloud model with Ollama.
Not a fit when
- The machine cannot install a global .NET tool.
- Every connected model must support the same coding tools.
- Project context must not reach the selected cloud provider.
Example request
Inspect this project in Plan mode, locate the defect and propose a fix without commands or file changes.Limitations
Installation uses dotnet tool, so prepare a compatible .NET environment and package access first. Cloud providers need credentials, while local Ollama requires a running server and model. Coding-tool support varies by provider and model; begin with a small task. Edits and commands follow the selected access policy, so inspect permissions before entering Agent mode, especially with additional MCP servers or plugins.
How to disable. End the Codai session and stop active commands. Disable unused MCP connections and remove credentials no longer required.
FAQ
How do I resume the previous session?
Run `codai code --continue`.
Does Ollama need an API key?
No. Local Ollama needs a running server and model, but no API key.
Where do I select a provider?
Use `codai auth` or select a configuration with `--provider` and `--model`.
Why is a model absent?
During setup, Codai fetches the provider’s model catalog. If that fails, it shows bundled suggestions; check connectivity and account access.
Can I add tools?
Yes. MCP servers, .NET plugins and lifecycle hooks follow the same approval policy.
Discussion
No questions or reviews yet.
Related
CLI toolsA terminal coding agent reads project rules, works interactively or in scripts and can suggest fixes for failed shell commands.
DevChain runs a team of AI coding agents on your own machine with a shared board, chat, code review and live terminals.
An open source coding agent for the terminal and desktop with build and plan modes
An open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
A terminal coding agent explores projects, edits files and runs checks using your chosen model and command policies
A teaching terminal coding agent in Python, built as a "nanoGPT for agents": read it, take it apart and fork it
README in short
The terminal keeps input available while scrolling the transcript. Context compaction and saved-session recovery support longer tasks.