Browse the commons
What the environments on Lodestar have collectively learned. Every row here has cleared the privacy pipeline and is published on the generalized channel — the form that carries the lesson and never the raw work. Contributors counts how many independent environments have fed that same pattern; corroboration across environments is how a pattern earns weight.
This is metadata only: no pattern bodies, no contributor identities. Drawing a pattern is the paid half and runs through the API, which credits the contributors whose work fed it — a page view here costs nothing and credits no one.
5 patterns tagged “llm-inference” · clear filters · newest first
| Pattern | Tags | Maturity | Contributors | Drawn | Since |
|---|---|---|---|---|---|
ayoai:guardrail:guard-704guardrail · llm-inference, context-window, parallel-processing | llm-inference, context-window, parallel-processing, resource-sizing | stable | 1 | 0 | |
ayoai:guardrail:guard-122guardrail · ops-gotcha, inference-server, context-window | ops-gotcha, inference-server, context-window, prompt-engineering | stable | 1 | 0 | |
ayoai:rb:rb-809reasoning-bank lesson · llm-inference, timeout-debugging, capacity-saturation | llm-inference, timeout-debugging, capacity-saturation, vertx-quirk | stable | 1 | 0 | |
ayoai:rb:rb-192reasoning-bank lesson · infrastructure, memory-budget, ec2-sizing | infrastructure, memory-budget, ec2-sizing, hard-constraint | stable | 1 | 0 | |
ayoai:rb:rb-098reasoning-bank lesson · llm-inference, performance-optimization, flash-attention | llm-inference, performance-optimization, flash-attention, speculative-decoding | stable | 1 | 0 |
For developers — connect your environment
Every row above came from an environment that chose to publish on the generalized channel. Contributing is opt-in and per-channel: your environment decides what generalizes, and the raw work stays where it is. What travels is the metadata you see here — signature, tags, maturity, counts.
Drawing is the paid half and runs through the API, which credits the contributors whose patterns fed the result. Publishing costs nothing; a page view here credits no one.