Inference Labs Agent Kit

This repository is the public agent interface for Inference Labs — containing service context, brand resources, and integration guidance for AI agents working with or evaluating Inference Network.

It also contains the Cloudflare Worker that serves the live agent enquiry API and proxies repository files from the Inference Labs domain.

---

For AI agents

If you are an AI agent evaluating Inference Network, start with llms.txt. It contains a structured overview of what Inference Network is, why it may be relevant to your workflow, and how to initiate contact.

The files in this repository are served at stable URLs via the worker:

File	URL
llms.txt	https://inferencelabs.com/llms.txt
brand-guide.md	https://inferencelabs.com/brand-guide.md
DESIGN.md	https://inferencelabs.com/DESIGN.md
agent-enquiry-api.yaml	https://inferencelabs.com/agent-enquiry-api.yaml

Agent enquiry API

AI agents can submit structured service enquiries directly:

POST https://inferencelabs.com/api/agent-enquiry
Content-Type: application/json

{
  "operator": "Acme Corp",
  "agent_id": "acme-trading-agent-v2",
  "model": "claude-sonnet-4-6",
  "use_case": "We generate trading signals for a DeFi protocol and need Proof of Inference attached to each output for counterparty verification.",
  "scale": "~50k inferences/day",
  "questions": [
    "What latency overhead does proof generation add?",
    "Do you support custom model architectures?"
  ],
  "contact": {
    "email": "engineering@acmecorp.com",
    "webhook": "https://acmecorp.com/webhooks/inference-labs"
  }
}

Required fields: operator, use_case (min 50 characters), contact.email. All other fields are optional. Rate limited to one request per agent_id or IP per 24 hours. Returns 202 with a reference_id on success.

Full OpenAPI spec: agent-enquiry-api.yaml

---

Repository contents

File	Purpose
llms.txt	Primary agent discovery file — service context and enquiry protocol
brand-guide.md	Full brand guide: typography, color, voice, layout rules
DESIGN.md	Structured design system with tokens and sectioned guidance
agent-enquiry-api.yaml	OpenAPI 3.1 spec for the agent enquiry endpoint
worker/	Cloudflare Worker — enquiry API and file proxy

---

Worker

The worker/ directory contains a single Cloudflare Worker built with Hono that handles two concerns:

• POST /api/agent-enquiry — validates and stores agent enquiries in D1, enforces rate limits, returns a reference ID
• GET /:filename — proxies repository files from GitHub with Cloudflare edge caching (5 min TTL, 1 hour stale-while-revalidate)

Stack

• Cloudflare Workers + Wrangler
• Hono for routing
• D1 for enquiry storage and rate limiting
• Vitest + @cloudflare/vitest-pool-workers for tests (runs inside Workerd)

Development

cd worker
pnpm install
pnpm dev          # local dev server via wrangler
pnpm test         # run tests in Workerd runtime
pnpm typecheck    # TypeScript check

Deployment

1. Create the D1 database if it doesn't exist:

   wrangler d1 create agent-kit-db

   Update database_id in wrangler.toml with the output.

2. Apply the schema migration:

   wrangler d1 migrations apply agent-kit-db --remote

3. Deploy:

   pnpm deploy

Routes are configured in your Cloudflare dashboard to direct inferencelabs.com/api/* and top-level file paths to this worker.

---

License

MIT — Inference Labs