distill

You can automate pattern extraction from logs and memory by running a distillation cycle that reads previous outputs, auto-detects topics, and produces structured summaries autonomously. It compresses retrospectives and learnings into concise outputs without human input.

Topic detection during data distillation works by automatically scanning memory logs to identify new signals within retrospectives and learnings. It then generates structured distillations per topic or uses swarm-mode for broader parallel operations.

Yes, you can run parallel agents for memory distillation by using swarm-mode operations. This allows multiple agents to process different topics simultaneously, extracting patterns from the Oracle brain and logging structured outputs without human intervention.

The best way to compress vast memory into actionable summaries while keeping provenance is using an autonomous distillation workflow. It applies L1-L4 distillation levels to logs and learnings, writing structured outputs that preserve the original signal.

No dependencies are required to run autonomous pattern extraction on your data. The skill operates independently to read previous distillations, auto-detect topics, and log structured outputs to the Oracle MCP using Sonnet for writing.

Using different models like Haiku and Sonnet for logging and pattern distillation optimizes the workflow: Haiku efficiently gathers data from logs, while Sonnet handles the complex writing of structured distillations, ensuring high-quality autonomous memory compression.