# ManchesterHumans

> ManchesterHumans is a drop-in, OpenAI-compatible LLM API where you set the percentage of human oversight on every request. At 0% the response is generated fully by an AI agent; at 100% every word is written by a real person; anywhere in between is a hybrid. Based in Manchester, UK.

## Core proposition

You choose how human your AI is. One parameter (`human_oversight`, a value from 0 to 1) is added to every API call. We handle the routing, the human reviewers, the AI fallback, and the billing — and it all returns through the same chat-completions endpoint your existing OpenAI code already speaks.

## Oversight tiers

- **0% — Fully Autonomous.** Pure AI agents. Fastest, cheapest. Used for low-stakes, high-volume work.
- **25% — Spot Checks.** Humans audit a random sample. Used to catch drift early without slowing the pipeline down.
- **75% — Human in the Loop.** Every high-stakes reply is reviewed before it ships. Bots draft, humans approve.
- **100% — Human Written.** Every word from a real person. Zero hallucinations. The original Human API.

You can mix tiers across routes — e.g. legal queries at 100%, marketing copy at 25%, internal tools at 0% — all from a single endpoint.

## Pricing

Output cost scales linearly with the `human_oversight` parameter.

- Input tokens: £0.05 per 1,000 tokens (flat, all tiers).
- Output @ 0% oversight: £0.02 per 1,000 tokens.
- Output @ 50% oversight: £0.085 per 1,000 tokens.
- Output @ 100% oversight: £0.15 per 1,000 tokens.

A Professional tier adds vetted domain experts (Legal, Medical, Financial, Technical, Academic), per-route oversight policies, and SLA contracts; pricing is custom per engagement.

## How it works

1. POST to `https://manchesterhumans.com/v1/chat/completions` with your `human_oversight` level (0 to 1). Same JSON shape as OpenAI's chat-completions endpoint.
2. Our AI agent generates a response. At 100% oversight, we skip straight to a human.
3. Per your oversight setting, a qualified person audits, edits, or rewrites the reply from scratch.
4. You get the standard OpenAI chat-completion JSON back. Your code doesn't know the difference.

## OpenAI compatibility

The API speaks the exact same format as OpenAI's. Point the Python, Node, PHP, or Ruby OpenAI SDK at `https://manchesterhumans.com/v1` and use the model name `human`. Streaming via SSE is supported.

Example (Python):

```python
from openai import OpenAI
client = OpenAI(api_key="manchesterhumans-...", base_url="https://manchesterhumans.com/v1")
response = client.chat.completions.create(
    model="human",
    extra_body={"human_oversight": 0.5},
    messages=[{"role": "user", "content": "What are the key principles of contract law?"}]
)
```

## AI Consultancy

ManchesterHumans offers AI consultancy that scopes, builds, and runs agent systems with human-in-the-loop oversight baked in from day one. Every system we deliver is out-of-the-box compatible with the same `human_oversight` parameter on our public API — no glue layer, no bespoke integration.

Engagements typically cover:

- **Scope & Proof of Concept.** Discovery workshop, then a two-week POC with the oversight API wired in. We map oversight tiers (0% → 100%) to your actual risk surface so you only pay for human review where it earns its cost.
- **Agents, Skills & Intelligent Code.** Tool-using agents, skill routing, retrieval pipelines, structured outputs. Native to the OpenAI SDK and to our HITL endpoint.
- **Planning & Execution.** Phased rollout, oversight policy, evaluation harness, expert pool setup, and production handover with a steady-state SLA.

A common pattern: start at 100% human oversight while the agent learns the domain, then slide the parameter down as evaluations confirm it's safe to run more autonomously.

To discuss a project: contact@manchesterhumans.com.

## Free Ethical AI Toolkit

Alongside the API, ManchesterHumans publishes a free, open-source ethical-AI compliance toolkit. No sign-up required. The toolkit covers ten documents:

1. AI Policy
2. AI Register
3. Risk Register
4. Use Case Worksheet
5. Bias Audit Checklist
6. Model Card Template
7. Impact Assessment Framework
8. Transparency Protocol
9. Data Governance Playbook
10. AI Incident Response Plan

There is also a free AI Ethics Assistant prompt that walks teams through setting up responsible-AI practices end-to-end.

## Pages

- [Home](https://manchesterhumans.com/) — Product overview, oversight tiers, pricing, API reference, live demo, signup
- [Consultancy](https://manchesterhumans.com/#consultancy) — Scope, POC, agents/skills/intelligent code, planning, execution; HITL out of the box
- [Ethical AI Toolkit](https://manchesterhumans.com/toolkit.html) — The 10 compliance documents and AI ethics assistant prompt

## Contact

- Email: contact@manchesterhumans.com
- Based in: Manchester, United Kingdom