Clavis is the AI harness you run, not the one you build.
A self-hosted orchestration hub that turns models into a working team of agents. Install it, add a key, and the harness is already assembled.
Clavis is a self-hosted AI agentic orchestration hub. It orchestrates agents, runs workflows, manages skills, conducts research through live web browsing, runs scheduled tasks, generates content, and is controlled from a dashboard, a CLI, and a chat interface.
The whole harness, one install.
Every component an AI harness needs, included and connected. No assembly, no glue code, no separate vendors to stitch together.
76 agent structures
16 user-facing domain agents for engineering, research, content, finance, commerce, and video, plus 60 infrastructure agents handling routing, memory, and error resolution. Build custom agents in Agent Studio.
Visual workflows
A drag-and-drop builder with standard, conditional, parallel, and negotiation steps. Assign agents, add retry logic, attach quality checks. No code.
Six-layer memory
Full-text and semantic search, causal chains, and a versioned wiki with an inline diff viewer. Agents read and write the same knowledge base.
Scheduler
Configure recurring and one-time agent tasks. Optional overnight autonomy plans and runs work in waves, then grades the results by morning.
Native CRM
Contacts, companies, deals, and activities with hybrid search and four-tier per-record sharing. Agents can read and write CRM records mid-conversation.
Multi-channel chat
Run agents from a web dashboard, a CLI, or through Telegram, Discord, and Slack. One session model, one command set across every channel.
Service integrations
Connect GitHub, Notion, Slack, Google Workspace, Linear, and Brave Search through the MCP service layer so agents act inside your stack.
5-gate security
Sandboxed file and shell access with symlink-proof path control, web content sanitized against prompt injection, and every new skill gated behind review.
Multi-model routing
Dynamic routing picks the right model per task, a circuit breaker drops failing providers, and per-call latency, token, and cost tracking feeds budget alerts.
Self-hosted, on your terms.
Your infrastructure
Installs on macOS, Windows, or Docker with a setup script. Your data, your keys, and all agent activity stay on hardware you control.
Your model
Bring an API key, route through your Claude Pro or Max plan, or run fully offline and private on a local model with Ollama. No key required.
Your team
Two UX tiers, simplified and advanced, with roles and per-record sharing. Non-developers operate agents through the dashboard; engineers keep the CLI.
A harness that learns from how you work.
Self-learning
Clavis tracks outcome signals from every tool and feedback action. When a pattern proves out, it drafts a rule and runs it on probation before promoting it. Agent personas evolve as the system learns what works.
Self-improving orchestrator
Repeated corrections become staged proposals for new guidelines and skills. Every proposal is reviewed by an admin before anything installs. Nothing changes behind your back.
Error resolution
A domain expert agent proposes a fix, a QA critic reviews it, and the change applies with rollback. Safety rails keep critical files off limits.
Ensemble reasoning
Three modes for hard problems: consensus across agents, a single best specialist, or a domain answer paired with a QA critique.
Common questions
Do I need to be a developer to run Clavis?
Is Clavis self-hosted?
Which AI providers does it support?
What version is Clavis at?
Get Clavis running today.
The fastest way to understand a harness is to run one. Reach the Hit Network team for a walkthrough on your own setup.