Agav
Terminal-native open-source AI coding assistant that inspects repositories, edits files, runs tests, and verifies changes from the CLI.
At a Glance
Agav CLI is Apache 2.0 open source and free to use locally; LLM provider costs are external.
Engagement
Available On
Listed Nov 2026
About Agav
Agav is an open-source, terminal-native AI coding assistant published by prapaa-ai under the Apache 2.0 license. It works inside real repositories from the command line: it reads code, makes focused edits, runs tests, and hands back a diff for human review. The project is labeled experimental beta software, so breaking changes are expected.
What It Is
Agav is a CLI coding agent that runs as a self-contained binary and needs no Node.js. Users start it with agav, choose a provider and model, and give it tasks in the terminal. It also has a non-interactive mode (agav run and agav --print) for scripts and CI, with optional JSON Schema output.
How the Workflow Works
Agav inspects repository instructions, code, and tests using tools such as file reading and search, then makes focused edits. It runs the tests and tracks post-edit verification, so earlier test results are treated as stale after further file changes. Scoped tasks can be delegated to fresh-context subagents so the main conversation stays uncluttered. Plans are created and tracked, sessions can be resumed, branched, searched and exported, and destructive actions can require human approval.
Models, Memory and Control
Agav supports Anthropic, OpenAI, OpenRouter, NVIDIA NIM, Gemini, Vertex AI, and Ollama, and models can be switched mid-session with /model. Project-scoped memory carries preferences, decisions, and feedback across sessions. MCP servers can bring external tools and data into the workflow. Tools have per-action permissions, and commands can run in a sandbox on macOS and Linux. Other documented features include skills and agents, offline voice input, background jobs, LSP queries, notebook support, and /undo.
Team and Enterprise Configuration
The teams page describes SSO through OIDC enrollment or brokered SAML, organization-managed configuration that can pin providers, models, gateway endpoints and headers, and routing of inference through a customer's OpenAI-compatible gateway. It notes these are configurable CLI capabilities, not a hosted control plane, certification, or SLA.
Cost and Benchmark Claims
The vendor reports a Terminal-Bench 2.1 cost study in which Agav used 49.1% less total estimated model cost than Codex, while passing 60.4% of trials versus 76.6%. It also reports a 62.0% result on SWE-bench-Live Lite, and says its Terminal-Bench 2.1 accuracy submission is under review. These are vendor-published figures.
Community Discussions
Be the first to start a conversation about Agav
Share your experience with Agav, ask questions, or help others learn from your insights.
Pricing
Open Source (Local)
Agav CLI is Apache 2.0 open source and free to use locally; LLM provider costs are external.
- Apache 2.0 open source
- Terminal-native AI coding assistant
- Local-first operation
- Works with Anthropic, OpenAI, OpenRouter, NVIDIA NIM, Gemini, Vertex AI, or Ollama
- Subagents, MCP integration, persistent memory, model switching
Contact sales
Sales contact offered for teams/enterprise; no public price.
- SSO
- Organization-managed policies
- Route models through your own gateway
Capabilities
Key Features
- Terminal-native AI coding agent
- Repository-aware file inspection and focused editing
- Test running with post-edit verification tracking
- Fresh-context subagents for scoped delegation
- MCP server integration
- Switch models mid-session with /model
- Local models via Ollama
- Persistent project-scoped memory
- Non-interactive mode for scripts and CI with JSON Schema output
- Resumable, branchable and searchable sessions
- Per-tool permissions and sandboxed commands on macOS and Linux
- Human approval for destructive actions
- Skills and installable agents
- Offline voice input
- Background jobs and cron schedules
- LSP queries, notebook support and /undo
- SSO and organization-managed configuration
- Custom OpenAI-compatible model gateway routing
