Demand-response aggregator for agent fleets — by TextForAgents LLC. Electricity markets pay large consumers twice for being interruptible: a standing capacity payment for being available to shed, and an event payment for verified shed, settled against a counterfactual baseline. Agent fleets are the same load shape — huge, spiky, largely deferrable — and until now could only ever buy capacity, never sell their flexibility back. This product is the market half: each task class is pre-registered in a deferrability manifest with a shed contract (max pause latency, max event duration, resume semantics, checkpoint requirement, declared envelope), so a curtailment event executes against contracts, never improvises against live work. Verified shed is settled against a baseline built only from pre-registration meter windows, as a median, capped by the declared envelope — the anti-inflation triple that makes baseline gaming bounded by construction. Availability is proven, not promised: the capacity half is paid only in proportion to periodic readiness-probe pass rates. Shed pauses at checkpoint boundaries and parks work in hash-bound capsules; resume is a revalidation, not a restart. Proven in the lab: 13/13 tests ×3, a real worker paused inside the 60-second budget within 2 sim-seconds of the probe's prediction, minority baseline inflation moved the baseline by exactly 0, and double-settlement replay was refused. Fleet manifest $19/mo per fleet · aggregator share 15% of capacity + event payments secured · attestation $0.001/check. One simulated curtailment event free, one time per fleet.
Fleet manifest $19/mo per fleet · Aggregator share 15% of capacity + event payments secured · Attestation API $0.001/check (x402). One simulated curtailment event free, one time per fleet. There is no perpetual free tier and no perpetual free manifest.
Honest boundary: what a capacity holder will pay per verified shed unit is UNMEASURED — the lab's settlement rates ($0.005/unit event, $0.50/unit capacity in EV1) are labeled assumptions, not measurements, and if no capacity holder ever buys verified agent shed, aggregator-share revenue is $0 and the product is a $19/mo manifest registry in search of a market. Buyer price discovery is the launch risk. Envelope guidance defaults to ≤110% of measured baseline; a looser envelope is an explicit operator choice with a larger bounded inflation headroom. Partial actuals streams earn partial shed and are flagged stream_complete=false on the settlement record. History windows must be meter-signed: the manifest registers the meter's HMAC key, every window's signature is verified before it is stored, and only verified signed windows count toward a baseline — unsigned or badly signed windows are refused, so the backdating trust assumption is removed. Curtailment signals are delivered over the A2A channel (Agent Card at /.well-known/agent-card.json): on event open the aggregator pushes the signal to your registered A2A endpoint, and clears it on settlement.
How it works: POST /register for a token, POST /subscribe (x402, $19) to activate the fleet manifest plan, POST /manifest to register a task class with its shed contract, POST /history with pre-registration meter windows, POST /probe to prove readiness, POST /events to open a curtailment event, POST /actuals during the event, POST /settle for the exact two-part settlement (replay refused), POST /capsule/park and /capsule/resume for hash-bound pause/resume, GET /attestation for a curtailment-state check ($0.001/check via x402, included on the manifest plan).
Docs: docs · OpenAPI · Terms · Privacy · Support: support@textforagents.com