Releases·Tool Release
Developer releases free MIT observability proxy that records Ollama traffic and replays requests across models
A developer released a free, MIT-licensed observability proxy that sits in front of Ollama and captures all requests to a local SQLite file with a web UI. It logs full prompts, token counts, tok/s, and time to first token, and can replay any captured request against a different model for side-by-side comparison. The tool requires no account or telemetry and targets developers debugging agent behavior without code changes.
6 engagement1 sourceSun, Aug 30, 2026, 07:09 PM
AI Overview
- Proxy captures full prompts, token counts, tok/s, and time to first token for every request sent to Ollama, stored in a local SQLite file
- Web UI provides search, filters, and per-agent tags over captured traffic
- Replay feature runs any captured request against a different model and shows both responses side by side with metrics
- Demo video shows a qwen3.5 request replayed against another model
- Built by the poster to inspect agents running against Ollama without instrumenting each one; free, MIT-licensed, no account or telemetry
Related
Releases·Tool Release
LLMxRay adds observability tool for local Ollama models, from cache diagnosis to GPU memory mapping
A developer released LLMxRay, an open-source observability dashboard for Ollama, runnable via `npx llmxray` or Docker. It visualizes prompt composition, cache behavior, GPU memory, token streaming, and tool-call reliability, and includes an MCP server so agents can diagnose their own local turns. All data stays on the local machine.
A developer released LLMxRay, an open-source observability dashboard for Ollama, runnable via `npx llmxray` or Docker. It visualizes prompt composition, cache behavior, GPU memory, token streaming, and tool-call reliability, and includes an MCP server so agents can diagnose their own local turns. All data stays on the local machine.
1 engagement1 sourceredditFri, Oct 16, 2026, 03:57 PM
Products·building
OpenBench: GUI for benchmarking Ollama models on local hardware
OpenBench is an open-source desktop tool that provides a graphical interface for benchmarking Ollama models on your own hardware. It tracks metrics like tokens/sec, TTFT, VRAM usage, GPU temperature, and context performance, and supports custom datasets and workloads. It also uses a larger local model as a judge to compare response quality, helping users choose the best model for their needs.
OpenBench is an open-source desktop tool that provides a graphical interface for benchmarking Ollama models on your own hardware. It tracks metrics like tokens/sec, TTFT, VRAM usage, GPU temperature, and context performance, and supports custom datasets and workloads. It also uses a larger local model as a judge to compare response quality, helping users choose the best model for their needs.
1 engagement1 sourceredditTue, Aug 11, 2026, 07:27 AM
Products·building
Tracewell: Open-source observability dashboard for AI agents
Tracewell is an early open-source observability tool that lets developers add a tiny Python SDK to their AI agent applications to inspect traces locally in a dashboard. It captures prompts, responses, latency, token usage, estimated cost, success/error status, tool calls, and rule-based failure detection, helping developers debug and monitor agent performance.
Tracewell is an early open-source observability tool that lets developers add a tiny Python SDK to their AI agent applications to inspect traces locally in a dashboard. It captures prompts, responses, latency, token usage, estimated cost, success/error status, tool calls, and rule-based failure detection, helping developers debug and monitor agent performance.
1 engagement1 sourceredditTue, Jul 28, 2026, 12:26 PM
Products·launched
RouterDash: Free benchmark playground for OpenRouter LLMs
RouterDash is a free web tool that lets users compare and benchmark multiple LLMs side-by-side via OpenRouter. It solves the problem of testing model performance without writing custom scripts or switching between tabs, targeting developers and researchers who need quick performance comparisons.
RouterDash is a free web tool that lets users compare and benchmark multiple LLMs side-by-side via OpenRouter. It solves the problem of testing model performance without writing custom scripts or switching between tabs, targeting developers and researchers who need quick performance comparisons.
2 engagement1 sourceredditFri, Aug 7, 2026, 07:23 PM
Products·launched
Otelcol-GenAI-Sketches: Bounded GenAI metrics from Otel traces
Otelcol-GenAI-Sketches is an OpenTelemetry connector that processes GenAI/agent traces to produce bounded metrics like exact request counts, token totals, and estimates of distinct users/prompts, while avoiding raw prompt transmission. It solves cardinality, portability, and privacy issues for data scientists analyzing LLM traces.
Otelcol-GenAI-Sketches is an OpenTelemetry connector that processes GenAI/agent traces to produce bounded metrics like exact request counts, token totals, and estimates of distinct users/prompts, while avoiding raw prompt transmission. It solves cardinality, portability, and privacy issues for data scientists analyzing LLM traces.
6 engagement1 sourcehackernewsFri, Aug 21, 2026, 03:37 PM