Deyad CLI

Local AI coding agent for the terminal — powered by Ollama. No cloud APIs, no telemetry, runs entirely on your machine.

Quickstart

npm install -g deyad-cli

Make sure Ollama is running with at least one model pulled:

ollama pull gemma4:26b

Then run:

deyad                              # Interactive REPL
deyad "add a login page"           # One-shot mode
deyad --print "explain this repo"  # Print response and exit
deyad --auto "refactor utils"      # Full-auto sandbox mode

Features

• 26 built-in tools + MCP server support
• RAG-powered context — BM25 codebase indexing
• Auto-lint feedback — lints changed files, self-corrects
• Sandbox mode — temporary git branch isolation
• Undo/rollback — git-based snapshots
• Session persistence — resume conversations across restarts
• Native tool calling — Gemma 4, Llama 3.2+
• Multiple modes — REPL, one-shot, --print, --auto
• Multimodal — attach images for vision models
• MCP servers — extend with external tools via .deyad.json

Performance Tuning (GPU / Fast Startup)

See the full guide in deyad-cli/README.md Performance Tuning.

Quick summary:

1. Set OLLAMA_FLASH_ATTENTION=1, OLLAMA_NUM_PARALLEL=1, OLLAMA_KEEP_ALIVE=-1 in the Ollama systemd service
2. (Optional) Create an ollama-warmup.service to auto-load your model on boot
3. (Optional) Set numThread / numGpu in ~/.deyad/config.json

Result: deyad --print "hello" in ~0.9s with a warm model.

Documentation

Full documentation, tool reference, slash commands, and CLI comparison tables are in:

• deyad-cli/README.md — complete CLI reference

Security Model & Permissions

Deyad lets you choose how much autonomy the agent receives via the --approval-mode flag:

| Mode | Agent may do without asking | Still requires approval | | ------------------------- | ----------------------------------------------- | --------------------------------------------------------------- | | Suggest (default) | Read any file in the repo | All file writes/patches, all shell commands | | Auto Edit | Read and write files | All shell commands | | Full Auto | Read/write files, execute shell commands | - |

In Full Auto every command is run network-disabled and confined to the current working directory for defense-in-depth. Deyad will warn if you start in auto-edit or full-auto while the directory is not tracked by Git.

Development

cd deyad-cli
npm install
npm run build
npm test
npm link    # makes 'deyad' available globally

Running tests

npm test                # full suite
npm run test:watch      # watch mode
npm run typecheck       # type-check without emitting
npm run lint:fix        # auto-fix lint + prettier

Contributing

We welcome contributions. Create a topic branch from main, keep changes focused, and run the full test/type/lint suite before pushing:

npm test && npm run lint && npm run typecheck

See the repository for contribution guidelines.

License

MIT