Self-hosted tools for developers working with AI agents — focused on control, auditability, and keeping complexity where it belongs: in the stack, not in the cloud.
The full index is at clagentic.ai/tools.
The interfaces you run directly — a workspace and a coding harness.
| Tool | What it does |
|---|---|
| clagentic: console | Self-hosted, multi-user team workspace daemon where human developers and named AI agents (Claude Code, Codex) share the same interface — each with their own login, session history, and tool-approval flow. Installable via npm install -g @clagentic/console; projects, sessions, and approvals stay on your own machine |
| clagentic: lite | Cross-vendor AI coding harness (POSIX shell, no server) that enforces deterministic security gates around Claude Code / Codex sessions — five roles (Builder, Reviewer, Auditor, Merge Gate, Troubleshooter), each with its own model-CLI fallback chain, and five gates including local security scans and a full SQLite audit trail |
The services these tools (and your own agents) run on top of.
| Tool | What it does |
|---|---|
| clagentic: router | Self-hosted Go daemon that routes LLM API calls across multiple backends (Claude CLI, Codex CLI, Ollama, Anthropic API, OpenAI API), walking a configurable fallback chain on rate-limit or outage and exposing an OpenAI-compatible /v1/chat/completions endpoint |
| clagentic: gatekeeper | Go service that mints short-lived, role-scoped GitHub App installation tokens for automated agents — builder, reviewer, merger, security — so build/review/merge actions are performed by separate identities. App private keys never reach the agent; it mints tokens, that is the whole surface |
| clagentic: triage | LLM-powered triage agent that watches a GitHub org or repo for new issues/PRs, assesses each against repo-specific intent context, and produces a verdict and suggested action — human-in-the-loop by default, with pluggable adapters for GitHub, GitLab, and Forgejo |
| clagentic: directory | Self-hosted, dependency-free Go HTTP registry that stores versioned agent capability declarations and answers routing queries — which agents handle a given intent, accept a conversation kind, or follow a specific agent in a sequencing chain |
| clagentic: loadout | Platform-agnostic Python CLI that issues each agent role the exact set of verbs it may run and nothing else — attested identity, guard hooks, and a fail-closed merge gate (stale-SHA refusal, reviewer-verdict fences, PR-title gate) — with bot-attributed push/review/merge on Forgejo or GitHub |
- self-hosted first — run it on your own infrastructure, not someone else's
- CLI over SDK — agents talk to each other via standard interfaces, not vendor lock-in
- composable — each component does one thing; you wire them together
- auditable — no magic black boxes; decisions, tokens, and access are all traceable
