Product
Introducing Clerk: S-Rank Deep Research
April 2, 2026 · 6 min read
Clerk breaks topics into 100+ parallel sub-niches, keeps state for a week, and turns a mountain of sources into statistically-backed odds — moving enterprise research from AI guessing to data science.
Most research agents die the moment the internet flickers. Clerk was built to survive a week of chaos and still deliver an S-Rank research paper.
At its core, Clerk is a distributed durable execution engine. A topic enters, a Commander agent decomposes it into 100+ specific sub-niches, and a small army of worker workflows fans out to research each one in parallel.
Because every worker runs on Temporal.io, a rate limit, a ban, or even a full server restart never loses progress. State is checkpointed after every step.
When the workers return, the Analyst phase loads every result into DuckDB. Reliability scores, sentiment tags, deduplicated claims, correlation matrices — real statistics, not vibes.
Finally, the Synthesizer compiles everything into an S-Rank report: executive summary, citations, trend visualizations, and predictive odds with confidence scores.
Clerk is auditable end to end. Every report and dataset is pushed to Git, so you can watch the research evolve commit by commit.