Remanence

Contextual memory that means AI doesn't start from scratch every session

A memory system for AI assistants that builds a knowledge graph passively from team work. Sharing context across developers and automatic progress saving translate into measurable reduction in token consumption and project orientation time.

−70%token reduction per session
$1,200monthly savings per developer
0data outside team infrastructure

Problem

AI coding assistants have no memory. Every session starts from scratch — the model scans the entire project, consumes tokens orienting itself in the structure, repeats discovery of the same dependencies and conventions. One developer's knowledge of the project is unavailable to another. Context is the most expensive resource in AI work — and it regularly vanishes.

Solution

Remanence listens to agent communication, passively builds a knowledge graph with relationships between concepts, and returns relevant context on demand. The agent sends text, receives precisely selected information — no explicit API, no developer-side configuration, no change in workflow.

GRAF WIEDZY — WIDOK ZESPOŁUAuthServiceUserRepoJWTPostgreSQLAPI GatewayRedisMigrationsEventBusLoggerOAuthCacheAktywna encja (częste użycie)Zanikająca encja (rzadkie użycie)Silna relacjaSłaba relacjaDev 1sesjaDev 2sesjaDev 3nowy

System architecture

Three-layer entity extraction
Aho-Corasick gazetteer for known concepts, Catalyst NER for people and organisations, LLM fallback for technical terms and acronyms.
Weighted knowledge graph
Entities connected by relationships whose weights grow with use. Context derives from graph relationships, not full-text search.
Deterministic retrieval
Identical input with identical graph state produces identical output. The LLM is not involved in the read path and introduces no non-determinism.
Knowledge lifecycle management
Unused information naturally decays. Unique facts — identifiers, keys, addresses — are automatically protected from deletion.
Privacy by architecture
The entire system runs locally on SQLite, with no external services beyond an optional LLM API. Data never leaves the developer's machine.
Team knowledge sharing
Project structure and convention knowledge available to the entire team. New developers receive full context from day one.

Impact on costs and productivity

Token consumption reduction
AI receives precise context instead of scanning the entire project. Smaller prompts translate directly into lower API costs.
Faster onboarding
New developers on a project receive full context immediately — team knowledge lives in the graph, not in documentation that quickly goes stale.
Elimination of repetitive work
AI stops rediscovering the same dependencies, conventions and architectural decisions from scratch every session.

Let's talk about optimising your team's AI costs

We'll show you the impact contextual memory has on token consumption and project orientation time for your developer team.

Book a call