Recallr
Persistent Memory Infrastructure for Human-AI Collaboration
Large language models are stateless. Every conversation begins from zero. Recallr changes this fundamental constraint by introducing temporal continuity to AI-assisted development — transforming ephemeral chat sessions into accumulated institutional knowledge.
The Problem We Solve
Today's AI assistants suffer from session amnesia. The decisions made, patterns discovered, and preferences learned in one session are lost when the context window closes. Developers repeatedly re-explain their coding style, re-establish project context, and re-teach preferences that the AI "knew" yesterday.
Recallr creates a memory layer that persists across sessions, enabling:
Preference Accumulation — Your prompts are preserved verbatim. Over time, your communication patterns, decision-making tendencies, and stylistic preferences emerge as learnable signals.
Knowledge Distillation — Raw conversation logs are compressed into structured knowledge artifacts: goals, approaches, decisions, and insights. What took 200 exchanges to discover becomes a 200-word summary instantly loadable into future context.
Experience Transfer — The lessons from debugging a Kafka consumer retry issue in January become available context when facing a similar RabbitMQ problem in March. Past experience informs future work.
How It Works
1. Session Consolidation
Copilot CLI stores session state in events.jsonl — a complete transcript of every prompt, response, and tool invocation. Recallr reads this data and uses the language model to extract what matters: the goal, the approach, the key decisions, and the learnings.
2. Preference Preservation
User prompts are stored verbatim, never summarized. These form a corpus of how you think, ask, and decide. This enables future systems to learn your style, not just your history.
3. Skill-Based Retrieval
A bundled skill teaches the AI how to access its own memory. When you say "recall the auth refactor" or "what did we decide about error handling", the skill searches the summary index and injects relevant context into the current conversation.
The Vision
Recallr is infrastructure for a future where AI assistants:
- Remember what you worked on together
- Learn your preferences from accumulated interaction patterns
- Improve by transferring insights across projects and time
- Personalize without requiring explicit configuration
This is the difference between a tool you use and a collaborator who knows your work.
Requirements
- VS Code 1.93.0+
- GitHub Copilot Chat
- GitHub Copilot CLI
License
Proprietary — All rights reserved.
