Local model discovery
Detect installed Ollama models, inspect their fit for your hardware, and assign defaults per agent.
Windows self-hosted AI workspace
HAICHI connects local Ollama models and optional cloud providers into one controllable system for agents, plans, discussions, and repeatable pipelines.
Local-first. Cloud-ready. No public server required.
One control surface
Build a working system of specialized agents, then see what they are doing, which model they use, and where every task stands.
Detect installed Ollama models, inspect their fit for your hardware, and assign defaults per agent.
Send work to one agent or dispatch several independent responses concurrently with bounded GPU usage.
Run parallel plan tasks, retry incomplete work, or pass output through a deterministic agent chain.
Let agents challenge and refine an approach turn by turn while keeping each response correlated.
Connect OpenAI, Anthropic, Gemini, or GitHub Models without making cloud access mandatory.
Keep plans, discussions, events, conversations, and memory in a local SQLite database.
A clear operating loop
Use Ollama locally or add an optional provider key from Settings. HAICHI discovers the available model list.
Give each agent a role, system prompt, provider, model, and carefully scoped capabilities.
Choose a channel, plan, discussion, or pipeline and override models only when the task needs it.
Review events and results, cancel active work, retry failures, and retain completed plan tasks.
Private by default
HAICHI is designed for a trusted Windows workstation, not a public multi-tenant SaaS. Localhost stays convenient while remote access can be restricted to a password-protected Tailscale connection.
Runtime policy
Fair and simple plans
Run entirely local on your own hardware or unlock advanced cloud orchestration.
$0/ forever
Great for getting started with private, local-first AI orchestration.
€49/ one-time
For developers who need maximum flexibility and cloud capabilities.
Custom
For organizations requiring team licensing and custom integrations.
FAQ
No. HAICHI can run entirely with local Ollama models. Cloud providers are optional and billed separately by those providers.
Yes. The default setup limits local inference to two concurrent agents and can reduce concurrency after a timeout or memory failure.
Yes. Run HAICHI in Tailscale mode with a unique remote password, then open the private Tailscale address from your phone.
Only when you explicitly enable the relevant capabilities. Dangerous tools and filesystem APIs start disabled.
No. HAICHI v1.1 is a self-hosted Windows product built for a single trusted workstation.
Windows release
Requires Windows 10 or 11. Ollama is recommended for local models. Tailscale is optional for private remote access.