ProAgent Me Docs

Payload caps, rate limits, per-task cost budgets, and IP protection that apply to every external-tool call.

External-tool calls pass through the same safeguards as web chat — plus a few extras to protect both you and us from runaway agentic loops (a caller that sends 30 ask calls in 60 seconds, for example).

Payload limits

Limit	Value	Why
`question` length	8,000 chars	Keeps per-call cost bounded. Agentic callers (Claude Code "review my codebase") should summarize, not paste whole files.
Total request body	20 KB	Defense in depth.
Response size	Unbounded	The agent's reply is capped by model output limits, not ours.

A 50 KB question returns HTTP 413 (or MCP error PAYLOAD_TOO_LARGE). The calling AI will retry with a smaller prompt.

Rate limits

Limit	Value
`ask` / `ask_agent` calls	5 per minute per user
`list_my_agents` / `search_marketplace`	No limit
`get_updates` / notifications	N/A (no polling endpoints in v1)

Hitting the rate limit returns RATE_LIMIT. Wait a few seconds and retry.

Per-task cost budget

Each conversation (conversationId) has a cumulative cost cap scaled by your subscription tier:

Tier	Cost budget per conversation
Free	$0.50
Starter	$2.00
Professional	$5.00

When a conversation exceeds its budget, the next ask call returns TASK_COST_BUDGET_EXCEEDED. The calling AI surfaces the error to you — start a new conversation (omit conversationId) to get a fresh budget.

The budget is specifically for agentic callers that iterate heavily. Normal back-and-forth conversations rarely approach even the Free-tier cap — typical per-call cost on an advanced-tier agent is ~$0.10.

Subscription quotas

External calls count against the same task and message quotas as the web chat. There's no separate pricing tier for API access:

New conversation (ask with no conversationId) → consumes 1 task slot.
Continued conversation → consumes 1 client message toward the per-task cap.
At tier limits, you get TASK_LIMIT or MESSAGE_LIMIT responses until the monthly period resets (or you upgrade).

See Plans & billing.

IP protection

The creator's knowledge base is never exposed through the API. Two guarantees are built into every call:

Responses are answer-only. Every reply passes through automated safety checks that refuse prompt-injection attacks and strip any accidental source-material leakage before the response reaches you.
The integration API has no file-serving surface. There is no endpoint, tool, or protocol hook that can return a creator's raw knowledge files, references, or Q&A. Attempts to request them fail before any AI call runs.

See IP protection for the full set of commitments and their contractual backing.

Escalation from external tools

Escalation to the human expert always happens in the ProAgent Me web app. The escalate tool returns a signed URL; payment and live chat with the creator happen there. See Escalation flow.

Limits & safeguards