@kognai/run

Run a Kognai agent from your sovereign vault. Reads the bootstrap files (SOUL.md, IDENTITY.md, USER.md, MEMORY.md) + the agent's prompt.md, assembles a system prompt, and executes the task against the model declared in the agent's agent.yaml.

# scaffold a vault first if you haven't
npx @kognai/init

# then run the included coder agent
echo "write a python hello world" | npx @kognai/run coder

Usage

npx @kognai/run <agent>                       # task from stdin
npx @kognai/run <agent> --task "<text>"       # task from CLI arg
npx @kognai/run <agent> --vault <path>        # use a non-default vault dir
npx @kognai/run <agent> --llm <provider/model># override agent.yaml's llm

| Flag | Default | Notes | |------------------|--------------|-------------------------------------------------| | --vault <path> | ./.kognai | Vault directory | | --task <text> | stdin | Task description | | --llm <p/m> | from yaml | e.g. ollama/qwen3:14b, anthropic/claude-haiku-4-5-20251001 | | --help, -h | — | Show usage |

Routing

agent.yaml's llm: field decides where the call goes:

• ollama/<model> → POST http://127.0.0.1:11434/api/chat (override host with OLLAMA_HOST)
• anthropic/<model> → POST Anthropic Messages API (requires ANTHROPIC_API_KEY)

If agent.yaml is missing or has no llm: field, defaults to ollama/qwen3:4b.

Environment

• OLLAMA_HOST — Ollama base URL (default http://127.0.0.1:11434)
• ANTHROPIC_API_KEY — required when llm provider is anthropic

The CLI auto-loads a .env file from your vault dir AND your current working directory (in that order). No dotenv dep required.

Zero npm dependencies

Raw HTTPS to Ollama / Anthropic. No @anthropic-ai/sdk, no axios, no js-yaml. The agent.yaml parser is a single-line llm: lookup; full yaml parsing isn't needed at runtime.

License

MIT