Q
What counts as positive feedback?
The served node appearing in successful tool calls, accepted edits, or passing tests downstream, recorded automatically per turn.
Glossary · Beta-Bernoulli Utility
Beta-Bernoulli Utility.
Score memories by whether they helped.
Direct answer
Each memory node carries a Beta(α, β) posterior over its usefulness; observed turn feedback updates it, and the mean û=α/(α+β) weights retrieval and packing. Nodes that actually helped get served more; unused ones fade even before decay. This closes the learning loop that flat files and additive memories lack: storage is not value, demonstrated help is.
In Anvaya
How We Implement It.
01Beta-Bernoulli UtilityAnvaya's Feedback service records TurnFeedback per served node; Gravitas multiplies outcome × decay × links × (0.5+0.5·û) at retrieval.
Questions
Asked About Beta-Bernoulli Utility.
Q
Why Beta-Bernoulli specifically?
Conjugate, online, interpretable: α/β count helps vs misses, the mean is a calibrated probability, uncertainty shrinks with evidence.
Q
Does utility override recency?
They multiply: a high-utility old node can outrank a fresh unproven one, exactly the senior-engineer behavior you want.
Run Agents That Fit On Your Laptop.
25.6 MB median RSS. 25 agents ran in parallel on a Core 2 Duo with 4 GB RAM. Hundreds on your machine. Zero cloud required on the Ollama path.
Requires Rust/cargo to build from source. Linux and macOS today, Windows not yet supported. Pre-1.0, public beta. Pricing TBD.