* Execute agent-browser CLI command. * Tries global agent-browser first, falls back to npx if ENOENT.

* Optimize attention computation using Flash Attention * @param input - Query, keys, and values for attention computation * @returns Optimized a

* Record a memory insight. * Stores in the in-memory buffer and optionally writes immediately.

* Run a shell command, return stdout or null on failure. * * SECURITY (audit_1776853149979): only call this with a STATIC command * string (no temp

* Run benchmark comparing Flash Attention vs baseline

* Store a learning trajectory for ReasoningBank

* Create a new rollout. Canary devices are selected as the first N% * of `targetDeviceIds` based on `policy.canaryPercentage`.

* Record a completed tool call. * Updates budgets and stores the result in the idempotency cache.

* Append a new event to the store

* Compute PageRank via power iteration with dangling node redistribution. * Returns map of node ID to PageRank score.

* Increment a counter

* Compute L2 norm of vector

* Resolve a SeedClient by device ID (or temporary endpoint key). * Throws when the device has not been registered.

* Save or update patterns. Existing patterns with matching IDs are * overwritten; new patterns are added. * * @returns The number of patterns

* Advance the rollout through its state machine: * pending -> canary (deploy to canary devices) * canary -> rolling (anomaly check p

* Build a minimal valid GGUF v3 binary buffer with the given KV pairs.

Cleanup resources

Clear cache

* Get team state

* Initialize the WASM module