Q
Why gate proactive memory?
Every unsolicited token competes with task tokens. EVOI keeps the precision high: speak only when expected value beats cost.
Glossary · EVOI Proactive Injection
EVOI Proactive Injection.
Memories that volunteer before they're asked.
Direct answer
Expected Value of Information gating decides whether an unrequested memory earns its tokens: inject only when utility × relevance-probability ÷ token-cost clears a threshold. Without a gate, proactive memory is spam — every injection taxes attention. With EVOI, the agent surfaces the past fix exactly when the current intent makes it likely-relevant, and stays silent otherwise. Proactivity with a budget.
In Anvaya
How We Implement It.
01EVOI Proactive InjectionAnvaya's Proactive service scores candidates as u·P(rel|intent)/tokens > τ each turn — memory volunteers under threshold control, measured by injection utility.
Questions
Asked About EVOI Proactive Injection.
Q
What feeds the relevance estimate?
Inferred session intent (bugfix, feature, refactor…) plus file, symbol, and recency signals — the same 9-signal relevance stack as retrieval.
Q
How is injection quality measured?
Mean injection utility (mind_stats): did injected nodes actually get used? The number should climb as the graph matures.
Stop Starting From Zero.
One binary. 11+9 Rust crates. 545 tests. Hand-written HNSW index. Three transport modes. Four providers, Ollama, Anthropic, OpenAI, Siemens. Zero API keys required to start. Mind remembers everything after the first session.