I got tired of juggling Ollama tabs, terminals, and agent chats

Running models locally is the easy part. Keeping multiple agents, prompts, plans, and project context organized is where my setup kept falling apart.

I would have Ollama running in one terminal, separate chats in different windows, notes in another file, and no clear answer to a basic question: which agent is doing what, with which model, inside which workspace?

So I built HAICHI, a self-hosted control room for local AI agent workflows on Windows and Linux.

The workflow I wanted

My minimum useful loop was:

1. Detect the Ollama models already installed on the machine.
2. Give two agents different roles and model assignments.
3. Run a plan, discussion, or sequential pipeline without rebuilding the context by hand.
4. Keep plans, messages, and results in a local SQLite database.
5. Allow file access only for explicitly selected models and workspace folders.

HAICHI v1.1 now does that from one local dashboard.

What the free build actually includes

The Personal build is free and does not require activation. It includes:

• Windows x64 and Linux x64 packages
• local Ollama model discovery
• up to two local agents
• one active workflow at a time
• plans, discussions, and pipelines
• persistent local state
• optional, workspace-scoped file permissions

The limits are deliberate. I want the first run to answer one question: is coordinating two local agents in one workspace more useful than juggling separate chats?

Developer Pro is a separate licensed build for cloud providers, unlimited agents and workflows, and private Tailscale access. You do not need Pro to test the local workflow.

Honest limitations

This is an early desktop product, not a hosted team SaaS.

• You need Ollama installed for local models.
• Tool reliability depends on the selected model following structured tool instructions.
• Smaller models can be useful for chat and planning but may ignore file-tool formats.
• File and command permissions are off by default because local automation should not receive broad access silently.

I am looking for three real first workflows

I do not need more generic launch impressions. I need three people who already use Ollama to try one real task and tell me exactly where they stop.

Try HAICHI Personal free: haichi.app

If download, installation, model discovery, or the first workflow blocks you, use the Tell us what blocked you link on the download section. Short and critical feedback is more useful than a polite review.

What is the first two-agent local workflow you would try: parallel research, code review, planning, or something else?