Limits and quotas

SignalMD has two kinds of limit. Plan credits follow your plan and count volume. Rate limits and sizes protect the service and are the same on every plan. This page lists both and what you see when you reach one. Pricing has the counting rules and the worked example.

Plan credits

	Free	Starter	Growth	Scale	Enterprise
Price per month, billed monthly	$0	$49	$179	$499	From $1,500
Price per month, billed yearly	$0	$39	$149	$425	Custom
Charged once a year	$0	$468	$1,788	$5,100	Custom
Credits a month	None	6,000	20,000	55,000	Set by contract
One-minute calls observed	None	6,000	20,000	55,000	Set by contract
Or about this many simulated tests	None	1,000	3,300	9,200	Set by contract
Or about this many copilot answers	None	300	1,000	2,800	Set by contract
Agents	1	3	10	25	Unlimited
Seats	2	3	10	20	Unlimited
Signed DPA	No	Yes	Yes	Yes	Yes

Every plan includes the board, flow, traces, versions, the REST API and the MCP server. Growth and Scale also include Slack support. Credits pay for processed calls, simulated and phone tests, and the copilot. The calls, tests and answers rows are each what the whole month’s credits buy if they are spent on that one thing alone, counting every call as one minute. Longer calls use more, and a month that mixes them gets a share of each. They reset each month and do not carry over. Prices are in US dollars. Agents and seats are totals. Paid plans cost less per month when billed yearly, and the yearly bill is charged once.

Allowance	What counts	At the limit
Credits	One monthly balance. Each processed minute of call audio uses 1 credit, and again when a call is analysed again. Each simulated test uses 6 credits for each test and persona in a run. A phone test uses 6 credits a minute, and at least 6. A copilot answer uses about 20. Quick checks and the brief’s Calibrate chat draw on the same balance	Calls keep importing and wait unanalysed, with no error. Tests and the copilot pause. Calls analysed again are read by rules alone. A run the balance does not cover is refused whole with a 402
Agents	Agents in the workspace now, not a monthly count. The demo agent is excluded	Connecting another is refused with a 402
Seats	People in your organization	Set by your plan. SignalMD does not count or remove members

A paid workspace’s credits reset at the start of each billing period, every month, annual plans included. Unused credits do not carry over. Free has no credits. Plan and usage under Settings shows the balance against its limit.

Plan limit errors

A request that the credits do not cover is refused with HTTP 402 and the code quota_exceeded. The API and the MCP server return the same fields as the app, in the API’s error object. error.quota says which limit, how much is used and which plan.

402 from POST /v1/agents/{agentId}/runs

{
  "error": {
    "type": "budget_error",
    "code": "quota_exceeded",
    "message": "This needs 72 credits and Starter has 40 of 6,000 left this month. Run fewer tests or upgrade.",
    "param": null,
    "requestId": "req_3f9a1c07d2b84e6a5c1d9b30",
    "quota": {
      "metric": "credits",
      "activity": "testing",
      "requested": 72,
      "limit": 6000,
      "used": 5960,
      "plan": "starter",
      "upgradePath": "/billing"
    }
  }
}

An MCP tool returns the same object as a tool error, with the status beside it:

Tool error from start_run

{
  "status": 402,
  "error": {
    "type": "budget_error",
    "code": "quota_exceeded",
    "message": "This needs 72 credits and Starter has 40 of 6,000 left this month. Run fewer tests or upgrade.",
    "param": null,
    "requestId": "req_3f9a1c07d2b84e6a5c1d9b30",
    "quota": { "metric": "credits", "activity": "testing", "requested": 72, "limit": 6000, "used": 5960, "plan": "starter", "upgradePath": "/billing" }
  }
}

Field	Meaning
`quota.metric`	The allowance that ran out: `credits` for a run or the copilot, or `agents` for connecting an agent
`quota.activity`	With `credits`: what the request would have spent them on: `observation`, `testing` or `chat`
`quota.requested`	With `credits`: the credits the request needed to start
`quota.limit`	What the plan includes for that metric
`quota.used`	What the workspace has used in the period (agents: the number connected)
`quota.plan`	The plan: `free_org`, `starter`, `growth`, `scale` or `enterprise`
`quota.upgradePath`	The app page to change plan: `/billing`, which is Plan and usage, under the app address

Through the API and the MCP server, the limit you can reach is a run the credits do not cover. Starting a run, or calibrating from a signed-in session, draws on credits. The copilot, the brief chat and connecting agents exist in the app only, and calls past your credits wait instead of failing. In the app, the same refusal shows as a notice with See plans. On Enterprise the message ends “Contact us to raise the limit.” Do not retry a 402 quota_exceeded as it is. Nothing changes until the month resets, you upgrade, or you start a run that needs fewer credits. Two other 402 codes are budgets and settings, not the plan:

budget_exceeded: your key’s monthly budget or the agent’s test budget did not cover the request.
testing_not_configured: testing is not set up for the agent.

API rate limits

Rate limits count requests in token buckets. A bucket holds the burst size and refills steadily at the rate, so after a quiet spell you can send a burst, then the sustained rate.

Limit	Applies to	Rate a minute	Burst
All requests	Each API key, or each signed-in person	600	120
Writes (a write route also draws from the all-requests bucket)	Each API key, or each signed-in person	60	20
Workspace	Every key, person and MCP call together, shared with the app	600	600
AI-backed routes (brief chat, test generation, persona drafts, rescore)	The workspace	10	10
MCP endpoint	Each client address	600	120
Failed authentication	Each client address, across the app, the API and MCP	30	30

Every /v1 response carries the tighter of the key’s buckets:

Header	Meaning
`RateLimit-Limit`	The bucket size
`RateLimit-Remaining`	Requests left in the bucket
`RateLimit-Reset`	Seconds until the bucket is full again
`Retry-After`	On a 429 only: seconds to wait before the next request
`Request-Id`	The id of this request, also in `error.requestId`. Quote it when you ask for help

A refused request returns HTTP 429 with the code rate_limited. A client address that has failed authentication too often gets too_many_failed_authentications instead, with or without a valid key, until its failures age out.

429

{
  "error": {
    "type": "rate_limit_error",
    "code": "rate_limited",
    "message": "Too many requests. Retry after 1 second(s).",
    "param": null,
    "requestId": "req_b41e7c90a2d85f3617e0c9a4"
  }
}

Wait for Retry-After, then retry. Treat the numbers as a guide to sustained use, not a guarantee. Calibrating tests needs a signed-in session. Every API key is refused with 403 and the code people_only.

Import limits

These are the limits on the hosted service. They bound how much one import brings in. They are not the plan credits.

Limit	Value
How far back an import reaches	30 days
Calls in one import	1,000
First import, on connecting the agent	The latest 500 calls from the last 30 days

After that, new calls arrive on their own, within minutes of ending, and are not held to the import size. You can keep only a percent of them, or cap them per day, week or month, in Import settings on Agent config. How many are analysed depends on the credits left. A call that runs past them is stored and waits. The import status on Agent config says how many are waiting, such as “120 analysed, 30 waiting for next month or an upgrade.” See Connect Retell.

Copilot limits

Limit	Value
Message length	8,000 characters
Answers at a time	One. Stop ends an answer early

A copilot answer uses about 20 credits, about any agent, the demo included. More lookups or a longer answer use more of them, and each quick check uses credits too. A question that needs too many lookups, a very long call or a very large lookup is cut short, and the copilot answers from what it read. Narrow the question if that happens. See Copilot overview for what the copilot reads.

Version comparisons

A comparison needs enough calls on each version before it gives a verdict. With too few, the measure reads Not enough calls and no verdict is given, however large the difference looks. Wait for more calls on the new version. An issue’s fix verdict also waits until the live version has taken calls for the agent’s grace period, 7 days by default. You can change it with Resolved after on the Issues page. See Versions.

Get started

Core concepts

Guides

AI copilot

Plans and billing

Reference

Limits and quotas

Plan credits

Plan limit errors

API rate limits

Import limits

Copilot limits

Version comparisons

​Plan credits

​Plan limit errors

​API rate limits

​Import limits

​Copilot limits

​Version comparisons

Plan credits

Plan limit errors

API rate limits

Import limits

Copilot limits

Version comparisons