Independent deployments will determine whether Yeschef can reliably dispatch Claude Code tasks across pooled LAN-hosted Ollama workers with useful throughput, task quality, and operational simplicity.
state: expiredheat: lowuncertainty: highknownscott: mediumcoding-agents local-inference agent-harnesses distributed-inferenceYeschefClaude CodeOllamaLabs Community
What is this?
Yeschef appears to be an agent harness that dispatches Claude Code tasks to a pool of Ollama workers on three LAN-connected NUCs, with the evidence title claiming aggregate throughput of 627 tokens per second. The supplied snippets establish that Claude Code can use Ollama-compatible backends and that developers are testing strong-model planner/local-model worker patterns to reduce costs, but they do not independently document Yeschef’s architecture or verify its throughput, task quality, reliability, or operational simplicity. Those claims therefore remain a single reported deployment awaiting replication.
Why it matters to Scott
The radar already tracks the core frontier-orchestrator/cheaper-worker claim in radar:multi-model-orchestrator-worker-agents, while commodity-machine pooling is covered by the Cascadia distributed-inference cases. Yeschef still bears directly on Scott’s active Ollama/LiteLLM routing stack and could provide a useful trace-backed replication fixture for testing whether headline throughput survives task-quality, reliability, and operating-complexity measurement, but the single reported deployment adds no validated result yet.
ip:framework.micro-agents-architectureip:concept.model-barbellip:source.subagents-speed-accuracy-ebookdev:technology.litellmdev:concept.task-aware-model-routingdev:technology.ollamadev:project.gamepcdev:concept.trace-backed-agent-comparisonradar:multi-model-orchestrator-worker-agentsradar:cascadia-distributed-intel-inferenceradar:concept.distributed-inferenceradar:concept.model-routing
queries asked of Scott's wikis
- planner-worker architectures for coding agents
- pooled local inference orchestration
- strong-model validation of local agent workers
- coding-agent throughput versus task quality
- LAN inference scheduling and fault tolerance
- operational simplicity of hybrid local-cloud agents
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (1) — ⭐ canonical anchor
Interpretation history
2026-08-27T16:31:21Z
The project has attracted no independent deployment, benchmark, or operational evidence after its initial claim; the lone additional comment does not change the case, which has faded as a standalone episode.
2026-08-25T15:50:39Z
No independent deployment, benchmark, or operational evidence has appeared; the initial repository and 627 tok/s claim remain an unvalidated single-deployment report. The unchanged observation cools this specific case despite continued activity in adjacent coding-agent and local-inference topics.
2026-08-25T15:37:32Z
grounded: known/medium — The radar already tracks the core frontier-orchestrator/cheaper-worker claim in radar:multi-model-orchestrator-worker-agents, while commodity-machine pooling is
2026-08-25T15:35:38Z
case created — The released repository implements a concrete hybrid frontier-agent and pooled-local-inference workflow with an initial throughput claim.
Decision trace
- 08-28 02:31expireThe project has attracted no independent deployment, benchmark, or operational evidence after its initial claim; the lone additional comment does not change the case, which has faded as a standalone e
- 08-28 02:31alert_silentThere is no consequential new delta to surface; independent trace-backed validation could justify a new episode if it later appears.
- 08-28 02:31alert_routeThere is no consequential new delta to surface; independent trace-backed validation could justify a new episode if it later appears.
- 08-26 01:50repriceNo independent deployment, benchmark, or operational evidence has appeared; the initial repository and 627 tok/s claim remain an unvalidated single-deployment report. The unchanged observation cools t
- 08-26 01:50alert_silentThere is no consequential new delta to surface; wait for an independent deployment or trace-backed evidence covering task quality, reliability, and operating complexity.
- 08-26 01:50alert_routeThere is no consequential new delta to surface; wait for an independent deployment or trace-backed evidence covering task quality, reliability, and operating complexity.
- 08-26 01:44alert_silentA single project post establishes that Yeschef is available and claims Claude Code dispatch across three LAN-hosted Ollama workers at 627 tok/s, but provides no visible evidence yet about task quality
- 08-26 01:44surface_candidateA single project post establishes that Yeschef is available and claims Claude Code dispatch across three LAN-hosted Ollama workers at 627 tok/s, but provides no visible evidence yet about task quality
- 08-26 01:44alert_routeA single project post establishes that Yeschef is available and claims Claude Code dispatch across three LAN-hosted Ollama workers at 627 tok/s, but provides no visible evidence yet about task quality
- 08-26 01:37groundThe radar already tracks the core frontier-orchestrator/cheaper-worker claim in radar:multi-model-orchestrator-worker-agents, while commodity-machine pooling is covered by the Cascadia distributed-inf
- 08-26 01:35createThe released repository implements a concrete hybrid frontier-agent and pooled-local-inference workflow with an initial throughput claim.