Where to look
Usage & Billing in the Portal shows balance, free credits remaining, current month spend, spending cap, workspace storage, and knowledge storage — plus top services, top capabilities, daily spend, and recent monthly statements. Per-agent and per-thread spend surfaces through the same usage data; the Overview page carries a spend-this-month tile and a runs-per-day chart.How tokens are metered
Prompt tokens are metered as three separate line items, because each carries its own rate:
Output tokens are metered separately. A call with no cache activity produces the
same two rows it always did — the extra buckets only appear when they are
non-zero.
Rates are per model and per route: the same model served over a different
route can carry different cache rates.
Non-token costs
Tool and capability usage is metered on its own units — per request, per second, per query, per hour, per GB-day. Web search is metered per query with a higher rate for deep searches. Computer is metered hourly, scaled by machine size. Slack and Telegram messages are free; SMS is not, and a provisioned phone number carries a monthly charge. Memory tool calls (memory_store, memory_recall, memory_forget) are not
billed as capability calls.
For current rates see noorle.com/pricing.
Platform limits
These are the request-shaped limits.
Rate limiting is a fixed 60-second window — one hour for SMS. There is no burst
allowance and no token bucket.
The two MCP gateway limiters fail open when Redis is unreachable: the request
is served rather than refused. The SMS limiter does the opposite — it fails
closed, and the message is not sent.
Noorle emits no
X-RateLimit-* headers. The MCP handshake 429 carries
Retry-After: 60; that is the only rate-limit header on the platform. On the
MCP surface a per-request rate-limit failure is a JSON-RPC error -32029, not an
HTTP 429.Resource quotas
Capability limits
Two per-account overrides exist for WASM plugins: a CPU fuel limit and a memory
limit, both clamped to the platform’s min and max. They are edited under
Settings → Limits.
Per-capability detail is on each capability’s page under
Built-in capabilities.
Balance and spending cap
The account’s spendable balance and its spending cap are enforced at the point a run or an MCP request allocates its budget. An MCP request with no budget available is rejected with 402 Payment Required. An agent run with no budget available does not start. The cap is checked against real period-to-date cost.Next
- Model routing — the biggest lever on token cost
- Autonomy and approvals