@utmiq/agent

Provider-agnostic AI coding agent CLI that encodes UTMIQ engineering standards. Runs against Anthropic, OpenAI, OpenRouter, Ollama, any OpenAI-compatible endpoint, or a built-in multi-provider router (OpenRouter-style routing, failover and aggregated model lists). Operates inside your repository through a permissioned tool system, and ships with Laravel-stack skills and standards.

Install

npm install -g @utmiq/agent
utmiq --version

Quickstart

utmiq init                 # write .utmiq/utmiq.db (SQLite) + rules
utmiq doctor               # check config, provider, credentials, git
utmiq                      # interactive session (multi-turn conversation)
utmiq ask "explain the billing flow"
utmiq run "add a failed_jobs index migration and its test"
utmiq serve                # local dashboard: usage + config editor

The interactive session keeps one conversation for the whole session: type a task, follow up naturally, use /help for meta commands, /model <id> to switch models upstreams, and press Ctrl+C to cancel a running task (then exit).

Set credentials via environment:

export ANTHROPIC_API_KEY=...     # provider: anthropic
export OPENAI_API_KEY=...        # provider: openai | openai-compatible
export OPENROUTER_API_KEY=...    # provider: openrouter

Commands

init, ask, run, review, fix, test, security, docs, skills, plugins, models, config, set, get, unset, list, doctor, serve, help. Running utmiq with no command starts an interactive session.

Configure without editing YAML (git-config style):

utmiq set provider.id anthropic
utmiq set provider.model claude-sonnet-4-5
utmiq set env.ANTHROPIC_API_KEY sk-ant-... --global
utmiq get provider.model
utmiq list --show-origin

Multi-provider router (like a local OpenRouter):

utmiq set provider.id router
utmiq set provider.strategy prefix
utmiq set provider.fallback openai
utmiq set provider.model openai/gpt-4o
utmiq set provider.routes '[{"id":"openai","model":"gpt-4o","apiKeyEnv":"OPENAI_API_KEY"},{"id":"anthropic","model":"claude-sonnet-4-5","apiKeyEnv":"ANTHROPIC_API_KEY"}]'

Exit codes: 0 success, 1 task/config failure, 2 validation failure, 3 approval required.

Configuration

Config lives in SQLite databases layered on top of built-in defaults: a global ~/.utmiq/utmiq.db and a project .utmiq/utmiq.db. Legacy config.yaml files are migrated automatically on first load. Permissions default to security-first (filesystem: project, shell: ask, mcp: ask, deploy/production: denied); dangerous commands always require approval and production requires UTMIQ_ALLOW_PRODUCTION=1.

Every agent run records usage (tokens, tool calls, model, provider, project) to ~/.utmiq/usage.db; disable with UTMIQ_USAGE=0.

See the project documentation for the full configuration reference.

Security

Repository content is treated as untrusted input. Secret-like files are never read and destructive commands are gated. Report vulnerabilities privately to [email protected].

License

MIT