requestintent preservedverbatim
decisionreasoning linkedordered
tool useactions retainedtyped
recoveryfailure path indexedtrainable
Marketplace access
The aggregate isfor members.
Sign in or create an account to view the live aggregate supply.
A living corpus for better agents
Training data compounds.Session by session.
Every accepted coding-agent session adds real requests, decisions, tool use, and recoveries to one credential-redacted training corpus.
Live aggregate supply
A corpus with
compounding depth.
Live supply needs JavaScript.
The collection, privacy, and publishing information below remains available.
Tradeable training tokens
—Token volume grows across longer, richer, real-world coding sessions.
Uploaded sessions
—Credential-redacted sessions
Active runtimes
—Native coding-agent sources
Useful data is behavioral
“The request matters.
The recovery teaches.”
Volume is only valuable when every token preserves what the agent tried, why it changed course, and how it reached a working result.
request
decision
tool use
recovery
From local session to corpus
Three steps.
One trusted path.
The public number is simple because the collection boundary is not.
-
01
local
Collect natively
Discover real coding-agent sessions through runtime-specific adapters on the contributor machine.
-
02
private
Redact before upload
Remove credentials and reject residual secrets before any candidate bytes cross the network boundary.
-
03
aggregate
Publish into the aggregate
Accepted sessions join one corpus and update the live totals without becoming public listings.
Add real behavior
Bring the sessions.
Grow the corpus.
Publish credential-redacted candidates with the Marketplace CLI. Every accepted session updates the live supply above.
trajectory marketplace seller candidate publish