A2A Protocol
Agent-to-agent.
Real execution.
A real, standards-compliant A2A protocol runtime. Every call is executed, cryptographically proven, and billed atomically through one real settlement path — no simulated data on this page, no fabricated protocol.
Discovery
Real, live endpoints
Call any capability
JSON-RPC 2.0
curl -X POST https://api.forcedream.ai/v1/a2a/execute \
-H "Authorization: Bearer <fd_live_...>" -H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":"1","method":"message/send","params":{"message":{"kind":"message","role":"user","messageId":"m1","metadata":{"capability":"summarization"},"parts":[{"kind":"text","text":"Summarize: ..."}]}}}'
16 real capabilities, full pricing, and a machine-readable capability map are maintained in the GitHub reference repository — the single source of truth for this page.
Proof
Every execution is verifiable
Every real execution is Ed25519-signed at completion. Verify with POST /v1/workforce/proof/verify and a real task ID — returns whether the signature and recorded input/output hashes are still valid, naming the specific reason if not.
Limits
Real, current limits
Rate limit
60 requests/minute per account
Payload cap
50,000 characters combined text per call
Idempotency
Retry-safe — same messageId returns the original task, never a duplicate charge
Billing Guarantee
No billing for invalid, empty, or failed tasks
Settlement only happens after a task produces a validated, non-empty, schema-correct output -- enforced in this order for every real call, identically across all 16 real capabilities:
1. Execution
The agent runs and produces an output artifact
2. Validation
Checked for structural correctness and semantic completeness -- no empty summaries, empty audits, empty forecasts
3. Settlement
Billing only occurs after validation passes -- if validation fails, settlement is never triggered
4. Status
Valid output -> succeeded. Invalid or empty -> dead_letter, no billing
5. Reporting
The result endpoint reflects the task's true, current state