Ollama agent workflow

Run local Ollama agents without losing the workflow.

Ollama makes local models practical. The hard part starts when a task needs roles, handoffs, retries, notes, and evidence spread across several tabs and terminals. HAICHI gives that local-agent work one visible workflow.

HAICHI local Ollama agent workspace with visible plan state and agent roles
One local workspace. Keep model selection, agent roles, workflow state, and results visible.

What this solves

Local AI work usually starts cleanly: one prompt, one model, one response. It gets messy when the result needs a second pass, a verifier, a file check, or a retry. HAICHI is built for that moment.

  • Assign each local agent a role, prompt, provider, model, and allowed capability set.
  • Use local Ollama models by default, then add optional providers only when you configure them.
  • Keep plans, handoffs, retries, and final output in one local workflow instead of separate chats.
  • Review what happened before trusting the result or changing project files.

A good first HAICHI task

Create two agents: a Reviewer and a Verifier. Give them one real project change. Ask the Reviewer for concrete risks and missing tests. Ask the Verifier to challenge every weak claim and keep only evidence-backed fixes.

This workflow is narrow on purpose. It makes HAICHI useful before you try bigger automation.