About
About stratless
stratless is a competitive-analysis API: give it a list of competitors as text, and get structured competitive data back — the axes they compete on, who clusters together, each one’s direct rivals, and the gaps — deterministic, calibrated, and traced to source.
Why it exists
AI made research instant — and mostly untrustworthy. An agent can scrape fifty competitor pages and produce a confident market map in a minute, and almost none of it can be checked. The analysis step is where the slop gets in: axes nobody derived, “rivals” nobody verified, verdicts nobody can trace.
stratless exists to make one step of that pipeline provable. It describes the field and refuses the verdict. It cites a source for every value, or says null. It returns the same map for the same set, every time. And when the input is a grab-bag, it stops and says so — it will not invent structure from noise. We would rather hand you a refusal than a confident guess.
The name is the mission: stratless — for the strategy-less, the people and agents working without a clear view of the field. We give sight, not a script. The map is yours to act on; the verdict is yours to make.
What it is — and what it refuses to be
stratless is the analysis layer of competitive intelligence — one operation family, done to a measured standard: classify a company from its own text, and analyze a set into a competitive field map. The boundaries are the identity:
- It describes; it never grades. No scores, no rankings, no “opportunity” — structure as fact, judgment left to you.
- It stores nothing. Your set is processed in memory and forgotten. No corpus, no retention — a design guarantee, not a marketing line (privacy).
- It returns JSON only. No prose, no reports. The output is built to be consumed by your code or your agent — the deliverable stays yours.
- It refuses incoherent sets. Companies that don’t form one field come back
coherent: falsewith the reason — never a fabricated map. - It stays in its lane. Not a scraper (bring any — Firecrawl, Jina, Exa, Tavily), not a database, not a CI suite, not a report generator. One station, done well.
How quality is proven
Trust isn’t asserted here; it’s measured. Every placement carries a confidence label, and the label is its measured accuracy: “high”-confidence placements matched expert-labeled reference maps 89.7% of the time, across 204 graded decisions on 3 labeled fields — a number we publish, re-measure with every release, and flag as provisional while the reference set grows. Determinism is part of the contract: same set in, same map out, byte for byte, so two runs are diffable. Fields the source text doesn’t support come back null — honesty about what the machine doesn’t know is a feature, not a failure state.
The numbers live on the homepage scorecard and the method lives in the docs — check both rather than taking this page’s word for it.
Who runs it
stratless is built and run by one person, from Malaysia. That’s a deliberate shape, not a stage: the product is designed to need no feeding — stateless, no corpus to maintain, no dashboard to babysit — so one careful operator can keep its quality bar for years. Small enough to answer every email; strict enough that you shouldn’t have to send one.
The code is on GitHub, the MCP server is on npm, and the person is at [email protected] — questions, bug reports, and “this placement is wrong” messages all welcome. The wrong-placement ones are the most valuable: they become the next labeled field in the reference set.