AxonGate

AxonGate answers one question for an autonomous caller: do the public sources actually support this claim, well enough to rely on? It fetches the live sources behind a claim, extracts clean context, and returns a citation-backed verdict — so an AI agent or RAG pipeline can decide whether to trust a statement before repeating it to a user.

It is a running commercial service, not a demo.

Live: api.axongate.one · interactive docs · OpenAPI contract · agent brief

Why the payment layer is the interesting part

Most APIs assume a human signed up, got a key, and attached a card. AxonGate assumes nobody did. An agent that has never seen this service before can discover it, learn the price, pay, and get an answer — in one request cycle, with no account, no key, and no human.

That is the x402 flow, and it works like this:

1. The agent POSTs to a paid route with no payment attached.
2. AxonGate replies 402 Payment Required with a machine-readable quote: the exact USDC amount, the receiving address, the chain (eip155:8453 — Base mainnet), and an expiry.
3. The agent signs a USDC transfer authorisation and retries with it in the X-PAYMENT header (PAYMENT-SIGNATURE and X-402-PAYMENT are also accepted).
4. A facilitator settles on Base; the response carries the work product.

Details worth noting if you are reading the code:

• The quote is selected per request, not fixed per route. x402_dynamic_price reads a tier or pack value from the request's query string or headers and resolves it against a price table configured by environment variables — roughly $0.012 to $0.05 for extraction tiers and $0.10 to $1.00 for Proof Packs.
• The service refuses work it would lose money on. Before doing paid work it computes projected profit as revenue minus live Base gas — the current base fee times a gas-unit estimate, converted through a live ETH/USD quote with a configured floor and cache — minus bounded supplier cost per attempt. If the projected margin falls to or below AXONGATE_PROFIT_MARGIN_USDC (default 0.01), the paid request is rejected with a payment validation error and ueg_rejections_total is incremented. A per-request unit-economics check matters more at this price point than at normal API prices: a few cents of revenue does not survive an unexamined gas spike.
• Two rails, one product. Agents pay per call over x402. Humans buying multi-source Proof Bundles pay through Stripe checkout. Webhook handling verifies an HMAC-SHA256 signature over the timestamped raw body with a constant-time compare, rejects timestamps outside a tolerance window, and de-duplicates by Stripe event ID in Redis so a replayed event cannot deliver twice. Both rails converge on the same evidence generation path.
• Attribution is built into the payment route. Paid routes are also mounted under POST /from/{source}/v1/x402/access, so the marketplace that sent a paying agent is recorded at settlement time rather than guessed from a referrer header.
• Free routes are genuinely free. Preview, sample, and quote endpoints make no paid supplier call and return supplier_spend: false, so an agent can evaluate the service before committing funds.

Endpoints at a glance

Route	Cost	Purpose
POST /v1/x402/access	paid (x402)	Paid web-to-markdown clean context extraction
POST /v1/x402/proof-pack	paid (x402)	Citation-backed evidence report
GET /v1/proof-pack/preview	free	No-spend mini preview
GET /v1/proof-pack/quote	free	Supplier-free price quote
GET /v1/proof-pack/reports/{id}/verify	free	Verify a retained report receipt
GET /manifest.json, /agent_manifest.json	free	Agent and payment discovery contracts
GET /health, /metrics	operator	Readiness and operational metrics

GET quote endpoints are preview-only and retain nothing. Create a resumable quote deliberately with POST /v1/quotes. The /operator and /metrics routes require AXONGATE_OPERATOR_TOKEN, sent as X-AxonGate-Operator-Token, a bearer token, or the browser-only operator_token query parameter.

Local setup

Prerequisites:

• Python 3.11+
• Node.js 20+ for the MCP and paid-buyer examples
• Railway CLI only when inspecting or deploying the production service

Create the Python environment:

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt

Install the JavaScript dependencies:

npm ci

Copy the environment template for local overrides. Never commit the resulting .env file.

cp .env.example .env

Copy-Item .env.example .env

The checked-in manifests contain the public AxonGate wallet address, so the app starts without private wallet credentials. Supplier, Redis, email, LLM, Stripe, and operator features stay disabled or use local fallbacks until their environment variables are configured.

Run locally

PORT=8000 uvicorn axongate_gateway:app --host 127.0.0.1 --port 8000 --reload

$env:PORT = "8000"
uvicorn axongate_gateway:app --host 127.0.0.1 --port $env:PORT --reload

Open http://127.0.0.1:8000/ for the product page, /docs for the service documentation, and /openapi.json for the agent contract.

Verify changes

The smoke suite uses an in-process ASGI client, fake Stripe signatures, and fake email delivery. It does not spend USDC or call paid suppliers. The x402 middleware may make a no-spend capability request to the configured facilitator.

python -m py_compile axongate_gateway.py examples/python_client.py scripts/ci_smoke.py scripts/secret_scan.py
python scripts/ci_smoke.py
python scripts/secret_scan.py

The same three steps run in CI on every push and pull request to main.

Production

Railway runs:

uvicorn axongate_gateway:app --host 0.0.0.0 --port $PORT

railway.toml configures /health as the deployment health check. The endpoint returns only service readiness information and also schedules throttled operational alert evaluation. Alerts are written as structured Railway logs by default and can additionally be sent to AXONGATE_ALERT_WEBHOOK_URL.

The production project contains the web and Redis services. Link a local checkout without changing production using the project ID from your Railway dashboard:

railway link --project <project-id> --environment production --service web
railway status

Do not run railway up, redeploy, restart, or variable mutation commands unless a production change is intentional.

Current structure

• axongate_gateway.py — FastAPI application, payments, evidence generation, delivery, analytics, and embedded UI
• scripts/ci_smoke.py — end-to-end in-process smoke suite
• scripts/secret_scan.py — tracked-file secret check
• examples/ — Python, MCP, curl, and paid-buyer examples
• manifest.json and agent_manifest.json — public agent and payment discovery contracts
• scripts/submit_x402_list.py — marketplace listing submission helper
• docs/ — custom-domain setup, marketplace distribution notes, and the listing submission kit

The application is currently a large monolith: axongate_gateway.py is ~20,500 lines, of which roughly a fifth is embedded HTML for the human-facing pages. New work should avoid adding more embedded UI or unrelated responsibilities to it; extracting billing, analytics, evidence generation, and page templates into modules is the current priority.

Licence

MIT — see LICENSE.