TaskQ gives builders one gateway for AI requests, queue jobs, policy routing, BYOK credentials, and HashIQ proof headers. Two environment variables. No silent fallback. No mystery execution.
Developers should not have to learn the whole execution topology to ship. The Gateway owns auth, policy, placement, proof, and settlement. You choose synchronous API calls or longer-running jobs.
Anthropic-shaped or OpenAI-shaped routes for synchronous completions, streaming, token count, and model listing.
Submit longer-running work that produces artifacts, reports, patches, generated files, and terminal job states.
Fetch the signed receipt for any request or job using the returned request ID or chain key.
POST /v1/messages POST /v1/messages/count_tokens GET /v1/models POST /v1/chat/completions # response headers x-taskq-request-id: req_8X3... x-taskq-chain-key: hshq_184467 x-taskq-placement-proof: proof_4b7...
POST /q/submit GET /q/thread/{job_id} GET /q/artifacts/{artifact_id} # terminal states queued → leased → running → succeeded queued → policy_blocked running → failed succeeded → settlement_reconciled
Developers can make execution requirements explicit. Hard constraints never silently relax. Empty compliant pools return a refusal with proof.
{
"provider": "anthropic",
"plane": "relay",
"must_run_in": ["CA"],
"must_not_run_in": ["US"],
"min_attestation_tier": "audited",
"key_custody": ["pinned", "enclave"],
"data_residency": "ephemeral",
"max_price_usd_per_mtok": 0.75,
"allow_queue_replacement": false
}Anthropic-shaped, OpenAI-shaped, and TaskQ-native routes let teams adopt proof without rewriting every workflow.
Tenant defaults, sub-keys, per-request policies, rate gates, budgets, and proof exports.
Every execution, refusal, credential access, artifact, and settlement event can map to the chain.
Two env vars, familiar endpoints, policy-bound routing, queue jobs, and HashIQ receipts.