HAICHILocal agent workspace

Reviewer + Verifier for local AI code review

One agent reviews. The second tries to prove it wrong.

HAICHI runs a local Reviewer and a local Verifier on your work. The Reviewer surfaces risks with evidence. The Verifier challenges weak claims. You keep the fixes that hold up.

  • Task entersPaste a change, describe a bug, or scope one local workflow.
  • Reviewer finds risksFailure modes, missing tests, and risky assumptions stay visible.
  • Verifier challengesUnsupported claims get discarded before you act.
Download Personal - free Try Reviewer + Verifier

Personal v1.2.0 · No activation · Windows 10/11 · Local Ollama · Up to 2 agents

HAICHI local run Reviewer + Verifier pipeline

Live workspace proof

Task queued Code change

Paste a diff, describe a bug, or scope one local workflow.

Reviewer running Finds risks

Failure modes, missing tests, and risky assumptions.

Verifier challenging Challenges claims

Rejects vague findings and keeps the evidence.

Evidence kept Fixes that hold up

You keep the next actions worth doing first.

HAICHI Command Center with local agents and inspectable workflow state

Runs locally in HAICHI. Real workspace state, agent roles, and handoffs stay inspectable.

A local review that earns your trust

  1. Task entersYou paste a diff, select files, or describe the change.
  2. Reviewer finds risksThe first agent surfaces issues with evidence.
  3. Verifier challengesThe second pass looks for gaps and weak claims.
  4. You keep what holdsEvidence-backed fixes are kept; weak claims are discarded.
206current source tests
1verified v1.2 Personal build
FreePersonal, no activation
Localplans and workspace state

Why HAICHI exists

Local AI chat is not the same as local agent workflow.

Chat windows and scripts are useful until context disappears, handoffs blur, and no one can tell which step produced the useful answer. HAICHI keeps the process visible.

Chat windows

No workflow memory

Good for one model conversation. Weak when a task needs roles, evidence, retries, and state.

Terminals + scripts

Powerful, easy to scatter

Logs, prompts, and outputs drift apart unless you build the workflow glue yourself.

HAICHI workflow

Roles, handoffs, retries

Reviewer and Verifier passes stay connected so you can inspect what happened before trusting it.

What the workflow looks like

Give the agents a task. Keep the work visible.

HAICHI does not hide a run behind one chat response. The plan, current state, agent handoffs, retries, and final output stay visible in the same local workspace.

Use cases

Searchable workflows, not vague AI promises.

HAICHI is easiest to judge when the workflow is concrete: local Ollama agents, code review, verifier passes, and a visible trail of what happened.

Ollama agents

Run local Ollama agents in one workflow

Move from scattered model tabs and terminal output to a visible local-agent workflow with roles, handoffs, and retries.

 Code review

Reviewer + Verifier for local AI code review

Use one agent to find concrete risks and a second agent to challenge weak claims before you decide what to fix.

 Windows

Run Ollama agents on a Windows workstation

Keep local models, agent roles, workflow state, and results together without turning your desktop into disconnected terminals.

 Coding workflow

Turn local AI coding into a repeatable workflow

Move from one-off prompts to a visible loop: task, review, verifier pass, fixes, and tests.

 Comparison

Open WebUI and HAICHI solve different local AI problems

Use chat UIs for model conversations. Use HAICHI when the work needs roles, handoffs, retries, and inspection.

 Reliability

Make local agent work visible before you trust it

Keep plans, decisions, retries, and final output inspectable so confident answers do not hide weak reasoning.

A concrete first run

One agent reviews. The second tries to prove it wrong.

Create a Reviewer and a Verifier with two local Ollama models. Give them one change from a real project, run the work as a plan, and keep both judgments visible instead of trusting one answer.

  1. ReviewerFind failure modes, missing tests, and risky assumptions.
  2. VerifierChallenge each claim and discard anything that is not supported.
  3. YouInspect both outputs and retry only the step that needs more work.
Starter task

Review this change for failure modes and missing tests. Reviewer: list concrete risks. Verifier: challenge each risk, discard weak claims, and return the three fixes worth doing first.

Watch the 38-second product tour
HAICHI collaboration workspace showing local agents and their visible work
The work stays inspectable. Agent roles, state, and output remain connected in one local workspace.

The useful parts in one place

The pieces I kept reaching for.

Models, agent roles, plans, handoffs, and results stay connected instead of being spread across separate tools.

01

Local model discovery

Find installed Ollama models and assign a default model to each agent.

02

Agent roles

Give each agent a focused role, prompt, model, and set of allowed capabilities.

03

Plans and pipelines

Run parallel plan tasks, retry incomplete work, or pass output through a deterministic agent chain.

04

Structured discussions

Let agents challenge and refine an approach turn by turn while keeping each response correlated.

05

Optional providers in Pro

Use local Ollama by default or connect supported cloud providers in the separate Pro build.

06

Workspace file permissions

Let selected models read or edit files only inside the workspace folders you approve.

A clear operating loop

From a task to an inspectable result.

  1. 1

    Connect models

    Use Ollama locally or add an optional provider key from Settings. HAICHI discovers the available model list.

  2. 2

    Assign agents

    Give each agent a role, system prompt, provider, model, and carefully scoped capabilities.

  3. 3

    Run the work

    Choose a plan, discussion, or pipeline and keep the task connected to its agent output.

  4. 4

    Inspect what happened

    Review events and results, cancel active work, and retry incomplete tasks without losing the thread.

Local by default

The control surface stays on your machine.

HAICHI is built for a trusted Windows or Linux workstation, not a public multi-tenant service. Developer Pro can add password-protected Tailscale access when you need a private remote connection.

  • Dangerous agent tools are disabled by default.
  • Agent file access is opt-in, model-scoped, and workspace-bound.
  • Provider keys remain in an ignored local credential file.
  • Cloud usage happens only for agents you explicitly configure.

Runtime policy

LocalhostDirect
TailscaleProtected
Public internetNot recommended
Dangerous toolsOff by default

Personal and Pro are separate builds

Start with the free local build.

Personal covers the core local workflow. Developer Pro is a separate licensed build for optional providers, more agents, and private remote access.

Founder test

€29 for the first 10 useful Pro users.

Use Personal first, run the Reviewer + Verifier workflow, then ask for a founder code before buying Pro. The normal public Pro checkout remains €49 unless a founder code is issued.

GitHub login should not block the request: copy the short text and send it as a reply or DM wherever you found HAICHI. The GitHub form is just the structured option.

Open request form
No activation

Personal v1.2.0

€0/ personal use

The free build for running a focused local workflow with Ollama.

  • Local Ollama model discovery
  • Up to 2 local agents
  • 1 active workflow at a time
  • Persistent local SQLite database
  • Plans, discussions, and pipelines
  • Personal, non-commercial use
Download Personal - free

Developer Pro v1.1

€49/ one-time

The separate licensed build for broader workflows and optional providers.

  • Everything in Personal
  • All 4 Cloud Providers (OpenAI, Anthropic, Gemini, GitHub)
  • Unlimited agents and workflows
  • Private Tailscale remote access
  • Perpetual use of HAICHI v1.x
Get Developer Pro v1.1

FAQ

Before you install.

Does HAICHI require paid AI APIs?

No. HAICHI can run entirely with local Ollama models. Cloud providers are optional and billed separately by those providers.

Are Personal and Pro separate downloads?

Yes. Personal v1.2.0 is the current free Windows build with no activation. Developer Pro remains a separate v1.1 licensed build with optional providers, more agents, and private Tailscale access.

Can I access it from my phone?

Yes with Developer Pro. Run HAICHI in Tailscale mode with a unique remote password, then open the private Tailscale address from your phone.

Can agents modify my files?

Only when you explicitly enable workspace-scoped permissions for selected models and folders. Write, delete, and command execution controls are marked as risk controls and start disabled. Tool reliability depends on the selected model following structured instructions, so stronger coder or cloud models are recommended for file-changing actions.

Is this a hosted SaaS product?

No. HAICHI Personal v1.2.0 is a self-hosted Windows product built for a single trusted workstation. Developer Pro remains on v1.1 until its activation flow is re-tested.

I am looking for useful feedback

Try HAICHI on one task you already do.

Personal is free with no activation and includes up to two local agents plus one active workflow. Start with the Reviewer + Verifier task above. If download, setup, model discovery, or the workflow gets in your way, tell me exactly where.

Download Personal - free Get Developer Pro v1.1 (€49) Open feedback form Personal v1.2.0 delivery through Lemon Squeezy. No payment required. SHA-256: 609c1691151b452a991609ec351b718a305f2e64b146630110d5326ecf048a22