TinyFish Documentation Audit
TinyFish's developer docs are unusually agent-friendly — llms.txt, llms-full.txt, four OpenAPI specs, a docs MCP server, and raw Markdown at .md paths — but the surface hides real cross-page contradictions: a limitations table that denies a flagship feature exists, an endpoint the reference refuses to list while three other pages depend on it, conflicting latency claims, and stale metadata that misleads the exact coding agents the docs court.
1. Browser Profiles "Limitations" table contradicts the Browser Context Profiles feature (significant)
Location: https://docs.tinyfish.ai/key-concepts/browser-profiles (Limitations) vs https://docs.tinyfish.ai/key-concepts/browser-context-profiles
Problem: The Browser Profiles limitations table states flatly: "Login persistence | Each run starts fresh, no session cookies carried over." But the Browser Context Profiles page is dedicated to the opposite claim: "Persist logged-in browser state across TinyFish runs… Browser Context Profiles let TinyFish save browser state — cookies, local storage, and session storage — and reuse it later." That same page's comparison table lists browser_profile: "stealth" alongside Browser Context Profiles as combinable capabilities, so the limitation isn't even true for stealth runs that attach a profile. The limitations row never scopes itself ("unless you use a Browser Context Profile") or links to the feature.
Consequence: A developer evaluating whether TinyFish can automate authenticated sites who lands on Browser Profiles first reads an unqualified statement that sessions cannot persist — and may abandon the product or build unnecessary per-run login flows. An AI agent consuming both pages receives two contradictory facts about session persistence and will assert whichever it read last.
The fix: Rescope the row to what's actually true (e.g. "Stealth alone does not persist sessions — attach a Browser Context Profile to carry cookies across runs") and link to /key-concepts/browser-context-profiles.
2. Agent API Reference omits /run-batch, which four other pages treat as a first-class endpoint (significant)
Location: https://docs.tinyfish.ai/agent-api/reference (Endpoints table)
Problem: The canonical Agent API Reference lists exactly three endpoints (/run, /run-async, /run-sse) and states "All three automation endpoints accept the same JSON body." Meanwhile the Webhooks page says "Webhooks are supported on all run endpoints: , , , and ", the Error Codes page documents behavior in three separate places (per-run validation envelopes, wallet-denial details, the pending-run cap), the FAQ recommends for bulk work, and llms.txt links a generated api-reference page for it ("Maximum 100 runs per request").