Every step is scored. The next tokens go to the work that earned them.
Prototype replay available. API is not open yet.
Replay of a scripted run · not live traffic
00 / 30
Budget spent and next-round share, by agent
Agent
Spent
Next round
planner
0.00
25.0% · 25,000
researcher
0.00
25.0% · 25,000
writer
0.00
25.0% · 25,000
reviewer
0.00
25.0% · 25,000
QUOKKA Pulse · Inbox clusters and the weekly digest budget 12.00 · 30 steps · fixture
Press Play. This is a recording, not a live model call.
Ledger 0 / 30
For
Agent platforms that already trace their runs and cap their spend, but still split the next round evenly. Pulse is the B2B API that sits on top: send each step in, get a score, a constrained reward, and the next round of budget. It works alongside your tracing and gateway, not instead of them.
Solves
A busy-looking fleet still burns budget. Repeat one high-scoring move and the eval is gamed. Busywork and polish loops spend the same tokens. When the run ends, nobody can show who actually contributed, so the next round is split evenly or by guesswork.
Does
Every step is scored, including handoffs and send-backs. The model only proposes quality and risk. Budget, anti-exploit, and novelty are decided by deterministic rules, so the raw score and the reward can differ — a high model score can still be cut. Each decision is written to a hash-chained ledger that can be recomputed from its inputs, and the next round of tokens follows that ledger.
The live endpoint is up. To verify:
$ curl -s -X POST https://www.quokkalabs.io/api/demo/pulse/run