Knowledge, Memory & Retrieval
Gestura does not rely on one giant blob of context. It separates live session context, durable memory, and a knowledge system so the right guidance can appear at the right time without overloading every prompt.
Three different sources of help
- Session context: the active conversation, approvals, and recent work.
- Durable memory: useful information preserved for later reuse across longer time horizons.
- Knowledge items: curated expertise such as Rust, Tauri, CLI, MCP, voice, and other focused domains.
What the knowledge system is for
The knowledge system is a progressive-disclosure layer. Instead of injecting every expert document into every request, Gestura can match relevant knowledge only when the current task benefits from it. That keeps normal prompts lean while still allowing deep built-in expertise.
Built-in and user-managed knowledge
Knowledge items can come from built-in expert documents or user-managed additions persisted on disk. Sessions can opt into specific expertise areas, and the CLI can help you inspect what knowledge is available.
How memory differs from knowledge
Knowledge is curated reference material. Memory is experience. Session memory helps Gestura stay coherent during active work, while durable memory lets important details survive compaction, long sessions, or later follow-up tasks.
How retrieval fits in
Retrieval is the step where Gestura decides what prior material to bring back into the current request. That can include recent session details, durable memory, or matched knowledge items. Good retrieval is what makes the system feel informed without making every response bloated.
When users should care
- You want the agent to benefit from built-in expert guidance.
- You need long-running work to survive context compaction.
- You want to inspect or search the knowledge system from the CLI.
- You are tuning how much prior context should come back into future work.