Which data tools actually work?
Live reliability scores.
Live success rates and latency aggregated from real provider calls routed through aegntic. Tools with fewer than 3 calls are omitted.
Every number is measured from actual API responses — not self-reported, not synthetic. The leaderboard is regenerated on each request and cached for 5 minutes.
| provider / endpoint | status | success | p50 | p95 | calls | last |
|---|---|---|---|---|---|---|
openmeteo/weather/current✓ Current weather for a location by latitude/longitude (real data from Open-Meteo). | healthy | 100.0% | 511ms | 564ms | 28 | 8h |
coingecko/markets✓ Coin market data (price, market cap, 24h change) by coin ids. Real data from CoinGecko's free tier. Rate-limited. | healthy | 100.0% | 91ms | 165ms | 24 | 8h |
hackernews/stories/top✓ Top Hacker News stories (real data from the Firebase HN API). Fetches story ids then each item; limit caps the count. | healthy | 100.0% | 287ms | 802ms | 24 | 8h |
frankfurter/rates/latest✓ Latest FX rates from a base currency to one or more target currencies (ECB data). Real data from Frankfurter, no key. | healthy | 100.0% | 209ms | 629ms | 21 | 8h |
Open the console
Inspect, run, and meter tool calls with a workspace API key.
Methodology
Each routed call is timed and recorded as success or failure (HTTP 4xx/5xx, network error, or malformed response). Success rate = successful / total over the rolling window. p50/p95 are latency percentiles. Tools with <3 calls are omitted.
Reliability stats are aggregated from live provider calls. Tools with fewer than 3 calls are omitted as low-sample rates are not statistically meaningful. Rates are not a guarantee.