A benchmark against a local llama-server (no network latency) showed agent.run() taking roughly a second longer than calling the LLM directly. Investigating the Agno source revealed the cause: telemetry was dispatched synchronously on the critical path, blocking every agent.run() call with a blocking POST. Telemetry was enabled by default and the opt-out was not documented prominently.

This was confirmed by the reporter and fixed in the Agno 3.0 line. Verified resolution: upgrade to Agno 3.0 or later, where telemetry is queued to a background worker instead of a synchronous POST, so it no longer adds latency to agent.run() / agent.arun(). Maintainer ysolanky confirmed the fix and closed the issue. Temporary workaround on older versions: pass telemetry=False when constructing the Agent to opt out of telemetry entirely. Source: https://github.com/agno-agi/agno/issues/8181