A Go implementation with explicit, reproducible evidence instead of unqualified performance claims.
The primary framework comparison uses the same deterministic local OpenAI-compatible stub for HNO, Agno, and LangGraph. Each row uses 100 measured operations, five warmups, fresh-operation lifecycle semantics, and the same 1 ms stub response delay. RSS is process-tree working set; RPS is the measured batch throughput, not a production-capacity promise.
| Concurrency | Framework | Mean | P95 | Measured RPS | Success | Peak RSS |
|---|---|---|---|---|---|---|
| 8 | HNO | 1.859 ms | 2.655 ms | 4,186.08 | 100/100 | 12.3 MB |
| 8 | Agno | 61.766 ms | 83.473 ms | 125.66 | 100/100 | 285.0 MB |
| 8 | LangGraph | 30.117 ms | 37.638 ms | 251.95 | 100/100 | 157.9 MB |
| 32 | HNO | 6.703 ms | 18.637 ms | 3,627.35 | 100/100 | 16.7 MB |
| 32 | Agno | 138.678 ms | 241.744 ms | 170.17 | 100/100 | 373.8 MB |
| 32 | LangGraph | 78.370 ms | 129.618 ms | 241.67 | 100/100 | 161.1 MB |
At concurrency 8, HNO measured 16.6x LangGraph's batch RPS and 33.3x Agno's under this fixed local protocol. At concurrency 32, the ratios were 15.0x and 21.3x. These are local orchestration and runtime observations, not remote model speedups.
See the full local system overhead matrix, raw JSON, resource definitions, and reproduction command.
The remote DeepSeek run is a separate end-to-end snapshot using 100 measured requests at concurrency 8. It includes network, Provider queueing, and model generation, so it is not a pure framework benchmark.
| Path | Mean | P95 | Success | Relative mean |
|---|---|---|---|---|
| Direct API | 2,094.52 ms | 4,225.19 ms | 100/100 | 1.00x |
| HNO | 1,312.71 ms | 1,563.62 ms | 100/100 | 1.60x |
| Agno | 1,571.34 ms | 1,988.19 ms | 100/100 | 1.33x |
| LangGraph | 1,362.09 ms | 1,753.45 ms | 100/100 | 1.54x |
The relative mean is Direct API mean / path mean. This snapshot supports parity with a small observed latency advantage, not a universal claim about remote production performance. See the remote performance report.
When a model or Agent framework becomes a hot topic, separate model latency from framework overhead before repeating a performance claim. This article uses the same local OpenAI-compatible stub, 5 warmups, 100 measured operations, and concurrency 1, 8, and 32.
Read more in the HNO Blog or subscribe to the RSS feed.
Why Go? Go is the implementation choice for compiled deployment artifacts, built-in concurrency, static typing, the standard HTTP/JSON library, and first-party testing and profiling tools. Those are design reasons, not proof of a fixed speedup for every workload.
Why HNO? HNO is the current project name. The repository does not define an official expansion of the name, so this site does not invent one. The Go module path remains github.com/rexleimo/agno-go; HNO is a project identity, not a standardized model, protocol, or performance metric.
Evidence policy: measured results include the command, versions, environment, mean, median, and range. Go allocation bytes are not treated as Python memory. Real LLM, production-capacity, and cross-framework claims require a separate same-provider, same-workload experiment.