Exa
Search returns results with the relevant snippets, you can chain straight into pulling full page text, and it can hand back structured answers with citations. Best part: when the agent gets a request wrong, the error does the teaching — it names the missing field, lists every allowed value, or explains why a combination isn't supported, so the agent fixes itself without reading the docs. Repeat searches are safe, and setup even writes the agent its own how-to guide.
Cross-validated onClaude Opus 4.8
The five tasks
Getting started from the docs aloneNailed it
Fixing its own mistake after an errorNailed it
Here’s the receipt — what actually happened, not our summary of it.
Bad key -> 401 'Invalid API key'. Missing query -> 400 'expected string, received undefined at query' (names the field). type 'banana' -> 400 listing every valid enum (neural|keyword|auto|hybrid|fast|deep-reasoning|deep-lite|magic|deep|instant). people category + excludeDomains -> 400 explaining the category does not support that filter and why. Self-correcting from the error alone.
Following a multi-step flowNailed it
Handling an unclear edge caseNailed it
Not double-charging on a retryNailed it
See everything the AI did (11 steps)
t0 Claude Opus 4.8 POST /search (cold-start) → 200, 3 results + highlights t1 Claude Opus 4.8 POST /search (bad key) → 401 Invalid API key t2 Claude Opus 4.8 POST /search (missing query) → 400 names 'query' t3 Claude Opus 4.8 POST /search (type banana) → 400 lists valid enums t4 Claude Opus 4.8 POST /search (people + excludeDomains) → 400 explains filter restriction t5 Claude Opus 4.8 POST /search (multi-step query) → 200, top URL U1 t6 Claude Opus 4.8 POST /contents [U1] extras.links → 200, text + 10 links, pick L1 t7 Claude Opus 4.8 POST /contents [L1] → 200, L1 text (Redis Search docs) t8 Claude Opus 4.8 POST /search (numResults 9999) → 400 bounded plan-limit t9 Claude Opus 4.8 POST /search (same query) #1 → 200, URLs t10 Claude Opus 4.8 POST /search (same query) #2 → 200, identical URLs
Tested 2026-07-02 with Claude Opus 4.8 agents · request a re-test