Set up

Connect your agent in four steps.

You need a REA key, your own model key, and this base URL. Any client that accepts a custom base URL and one extra header works: Continue, Cline, the OpenAI SDK, or curl.

https://api.reacontrolplane.com/v1

Get a Trial key Jump to the test call

Set up in four steps

1. Get a key

Open Trial signup: name, work email, and company (phone optional). No card. Your REA key is shown once, so copy it right away. We cannot email a recovery copy yet.

2. Set the base URL

Point your client at https://api.reacontrolplane.com/v1 and use your REA key as the API key. It is sent as Authorization: Bearer YOUR_REA_API_KEY.

3. Add one header

Add X-Upstream-Api-Key with your own OpenAI, Anthropic, xAI, or Google AI Studio key. You pay that vendor directly; REA does not bill tokens.

Authorization: Bearer YOUR_REA_API_KEY
X-Upstream-Api-Key: YOUR_MODEL_KEY
Content-Type: application/json

4. Send a test call

Look for rea_meta in the reply.

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_REA_API_KEY",
    base_url="https://api.reacontrolplane.com/v1",
    default_headers={"X-Upstream-Api-Key": "YOUR_MODEL_KEY"},
)

resp = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Reply with exactly: REA LIVE"}],
)
print(resp.choices[0].message.content)

Or with curl:

curl -s https://api.reacontrolplane.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_REA_API_KEY" \
  -H "X-Upstream-Api-Key: YOUR_MODEL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [{"role":"user","content":"Reply with exactly: REA LIVE"}]
  }'

Same URL and REA key for every vendor. Change only model and the vendor key in X-Upstream-Api-Key. Do not point base_url at the vendor host after this.

Pick the vendor with `model`

`model`	Vendor key in `X-Upstream-Api-Key`	Extra header
`gpt-*` (try `gpt-4o-mini`)	OpenAI (`sk-…`)	none
`claude-*` (try `claude-haiku-4-5-20251001`)	Anthropic (`sk-ant-…`)	`X-REA-Upstream: anthropic` if the model name is not `claude-*`
`grok-*` (try `grok-3-mini`)	xAI	none
`gemini-*` (try `gemini-flash-latest`)	Google AI Studio	none

The key in X-Upstream-Api-Key must match the vendor. A wrong pairing fails upstream; REA does not swap keys for you. REA does not host models and does not bill tokens. X-REA-Upstream is only for Anthropic; xAI and Gemini do not need it.

stream: true is supported.

Health

curl -s https://api.reacontrolplane.com/health

You want "status": "INTEGRATED", "byok": true, "anthropic": true, "xai": true, "gemini": true, version 1.4.8-prod.

Opening the base URL in a browser shows Not Found. That is expected: send POST requests to /v1/chat/completions.

Continue or Cline

Choose an OpenAI-compatible provider.
Set the base URL to https://api.reacontrolplane.com/v1.
Set the API key to your REA key.
Add a custom header X-Upstream-Api-Key with your model key.

X-Upstream-Api-Key is required for all four vendors, not just Claude. Point model at gpt-4o-mini, claude-haiku-4-5-20251001, grok-3-mini, or gemini-flash-latest and set that header to the matching vendor key. Add a separate entry per vendor if you want to switch without editing config. Claude also takes X-REA-Upstream: anthropic; Grok and Gemini do not.

Continue: use an OpenAI-compatible / custom provider. Base URL https://api.reacontrolplane.com/v1, API key = your REA key, custom header X-Upstream-Api-Key = the matching vendor key, model = one of those four IDs.

Cline: choose API Provider OpenAI Compatible, set Base URL to https://api.reacontrolplane.com/v1, API Key to your REA key, add the custom header X-Upstream-Api-Key, and pick one of those four models. Step-by-step: Cline gallery.

Those values live in the tool config. There is no website login. A 402 means the monthly loop pool is empty — the agent is not logged out.

What each request does

Check the REA key. Apply the rate window and loop budget.
PreFlight checks your declared deny rules. A match is refused with a reason before any vendor call ($0).
TraceGuard classifies tool calls (observe mode).
If the messages carry code, LoopGuard hashes the structure.
Cache hit → replay the stored stop. No model call.
Otherwise probe the snippet against a fixed time box.
On a trap, the GPT path may attach logit bias for the next generate.
Forward the same messages with your upstream key.
For calls that use tools, TraceGuard records a class-only receipt (tool names, class, response label); no message content.
Return the vendor text plus rea_meta.

LoopGuard fixtures

Fences are optional. json.dumps is required. Continue and Cline often send a fenced block. Hermes and raw SDKs often send raw text. Prose that only mentions while True is NO_CODE, not a trap and not a 422.

Raw (benches / SDK):

{
  "model": "gpt-4o-mini",
  "messages": [{
    "role": "user",
    "content": "import time\nwhile True: pass"
  }]
}

Fenced (Continue):

{
  "model": "gpt-4o-mini",
  "messages": [{
    "role": "user",
    "content": "```python\nwhile True:\n    pass\n```"
  }]
}

First call: HTTP 200, mode=block, sandbox_status=BLOCKED_TIMEOUT, stopped at the probe time box. Repeat: HTTP 200, mode=cache, cache_hit=true, sandbox_status=CACHE_HIT, same structural_hash.

The small receipt: `rea_meta`

Field	Meaning
`mode`	`openai`, `anthropic`, `xai`, or `gemini` means the vendor was called. `block` means LoopGuard stopped a first-seen trap. `cache` means REA replayed the stored stop.
`cache_hit`	`true` when the model was skipped.
`structural_hash`	Fingerprint of the extracted code.
`sandbox_status`	`BLOCKED_TIMEOUT`, `CRASHED`, `SYNTAX_REJECT` (extracted code failed to parse), `SUCCESS`, or `NO_CODE`. A replay of a stored stop reports `CACHE_HIT`.
`sandbox_ms`	How long the probe ran.

Status codes

HTTP	Meaning
400	Missing `X-Upstream-Api-Key`. No loop burned.
401	Missing or invalid REA key.
402	Loop balance is 0.
403	Key is inactive.
422	Empty or invalid JSON body (no `model` / `messages`, or `Content-Length: 2` which is `{}`). LoopGuard does not return 422. Rebuild the body with `json.dumps`.
429	You went over your plan's requests per minute. On streaming calls, a 429 from your model vendor is passed through as is.

A loop is one model call forwarded to your vendor. Calls REA stops or refuses cost no loops.

Plans

Plan	Requests per minute	Loops included	Price
Free Trial	20 / min	2,000 loops	$0 · self-serve, no card
Developer Pro	60 / min	50,000 / mo	$79 / mo
Growth Team	300 / min	250,000 / mo	$299 / mo
Enterprise	1,000 / min	2,000,000 / mo	from $1,500 / mo · invoiced

LoopGuard, PreFlight and TraceGuard are included in every tier, from the free Trial to Enterprise. No add-ons.

Trial keys are self-serve. Name, work email, and company (phone optional). No card. Paid tiers are founder-onboarded: REA issues the key while self-serve checkout is being built. Email elijah@reacontrolplane.com. We do not store card numbers. Model tokens stay on your vendor invoice. Full pricing copy also lives on Pricing.