Agav CLI
Agav is a terminal coding agent that reads a repository, edits files, runs tests and switches between seven model providers.
- 178 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The agent runs commands and edits files in a repository, and auto-accept mode removes confirmations.
- Code and prompts go to the chosen model provider unless local Ollama is used.
Install
Manual install
curl -fsSL https://agav.dev/install.sh | bashSelf-contained binary, no Node.js needed. For Windows PowerShell the README gives `irm https://www.agav.dev/install.ps1 | iex`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent runs commands and edits files in a repository, and auto-accept mode removes confirmations.
- Code and prompts go to the chosen model provider unless local Ollama is used.
This is third-party code. Review the repository files before installing.
What it does
Agav is a terminal AI agent for working in real repositories. It studies code, makes edits, runs tests and verifies the result from the command line. The model can be switched on the fly: Anthropic, OpenAI, OpenRouter, NVIDIA NIM, Gemini, Vertex AI and local Ollama are supported, switching happens inside a session with `/model`, and `/fast` and `/deep` give quick and thorough modes. Sessions can be resumed, branched, named, searched and exported, and `/compact` frees context while keeping plans. The `/permission` command switches between asking for confirmation, auto-accept and deny-writes. Scripts and CI use `agav run` and `agav --print` with JSON Schema output. Skills, fresh-context subagents and per-project memory across sessions extend the agent.
Who it is for. Developers who want a terminal coding agent with a choice of providers and local models through Ollama.
Good fit when
- You want to hand a repository task to an agent from the terminal and get edits together with a test run.
- The model has to change mid-session: a fast one for small things, a deep one for a hard spot.
- Code review runs in CI with a single `agav run` command and the answer is collected as JSON.
- Code cannot go to external providers, so the agent runs on a local model through Ollama.
- A long session has to continue tomorrow with the same plans and project memory.
Not a fit when
- You need stability: the project calls itself an experimental beta and breaking changes are possible.
- The team standardized on Claude Code or Codex and will not maintain a second agent.
- A company security policy forbids installing through `curl | bash`.
Example request
Start agav in the project root, switch to the `/deep` model and find why the test in `tests/auth` fails, proposing a fix without writing to disk.Limitations
The project calls itself an experimental beta, so breaking changes and rough edges are possible. You need a key for the chosen provider or local Ollama with a downloaded model. Command sandboxing is stated for macOS and Linux only, not Windows. The `/voice` input recognizes English speech only. Background jobs and schedules are opt-in and off by default. The installer script is downloaded and run from the network, so read it before running.
How to disable. Delete the `agav` binary the installer placed and the directory with settings and sessions. Revoke provider keys separately at the providers.
FAQ
How do I install Agav?
Run `curl -fsSL https://agav.dev/install.sh | bash`. The installer places a self-contained binary and needs no Node.js. Windows PowerShell has `irm https://www.agav.dev/install.ps1 | iex`. `agav update` keeps the agent current.
Which models and providers does Agav support?
Seven providers: Anthropic, OpenAI, OpenRouter, NVIDIA NIM, Gemini, Vertex AI and Ollama. A launch with a choice looks like `agav --provider ollama --model llama3.2`. Inside a session `/model` changes the model.
Can Agav run in CI without a dialog?
Yes. `agav run "review the code in src/"` and `agav -P` work non-interactively, tool permissions are set in advance, and the answer can follow a JSON Schema.
Does Agav ask before making edits?
The default mode asks. `/permission` switches it to auto-accept or deny-writes, and the choice persists when the session resumes. Destructive actions are stated to need human confirmation.
Where does Agav keep keys and data?
API keys are stored encrypted and a local mode is available through Ollama. The author states there is no hidden telemetry, and the code is open under Apache 2.0, so behavior can be audited.
Discussion
No questions or reviews yet.
Related
CLI toolsAn open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
A terminal coding agent in one Rust binary that reads code, edits files and runs tests with any model of your choice.
A teaching terminal coding agent in Python, built as a "nanoGPT for agents": read it, take it apart and fork it
Self-hosted AI coding agent with a web interface: project map, decision memory, local models via Ollama and permission modes
The AdaL terminal agent: switches models, shows edits and plans for review and hands the session over to its own IDE.
A multi-agent harness that verifies results by artifacts: stop hooks, an independent judge and an event log for Claude Code, Codex, Cursor and more
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Agav (this card) | no | 178 | |
| OpenCode | no | 226,389 | also a terminal agent with provider choice; Agav adds a non-interactive JSON Schema mode, project memory and background jobs |
| Crush | no | 29,067 | another author's terminal agent; Agav ships as a self-contained binary and splits permissions into ask, auto-accept and deny-writes |
README in short
One-command install without Node.js, `agav` to start and a non-interactive `agav run` for scripts. Details live in the project documentation site.