Use mnemo as Robin's persistent memory layer #17

New issue

Open

opened 2026-06-03 18:22:54 -07:00 by pyr0ball · 0 comments

pyr0ball commented

2026-06-03 18:22:54 -07:00

Owner

Summary

Robin is a proactive local LLM assistant for Linux migration. Its core value proposition requires persistent memory — remembering what the user has done, what problems they've hit, what their preferences and workflow patterns are. mnemo (https://github.com/zaydmulani09/mnemo, MIT) is exactly the right primitive for this.

What mnemo provides

Watches conversations, extracts named entities (apps, commands, file paths, errors, preferences) and relationships
Builds a persistent knowledge graph in SQLite
Injects relevant context back into future prompts automatically in under 50ms
Single static Rust binary, zero cloud dependency, Python SDK
Works with Ollama (Robin's default local LLM backend)

Robin-specific entity schema ideas

Apps: installed apps, equivalents found, ones the user likes/dislikes
Commands: commands the user has run, errors encountered, fixes applied
Preferences: DE/WM settings, theme, input devices, accessibility needs
Migration state: what has/hasn't been migrated from Windows/macOS
Problems: recurring issues and their resolutions

Architecture

Robin frontend → Robin backend (FastAPI) → mnemo sidecar (Rust)
                                         ↓
                                    SQLite knowledge graph
                                    (persists across sessions)

Why this matters for Robin's ND/adaptive user base

ND users often have inconsistent session patterns — they may not follow up on a problem for days. mnemo's persistent graph means Robin can pick up exactly where it left off without requiring the user to re-explain context. This is a significant accessibility win.

Linked

circuitforge-core: mnemo as cf-core memory module (shared implementation)

## Summary Robin is a proactive local LLM assistant for Linux migration. Its core value proposition requires persistent memory — remembering what the user has done, what problems they've hit, what their preferences and workflow patterns are. mnemo (https://github.com/zaydmulani09/mnemo, MIT) is exactly the right primitive for this. ## What mnemo provides - Watches conversations, extracts named entities (apps, commands, file paths, errors, preferences) and relationships - Builds a persistent knowledge graph in SQLite - Injects relevant context back into future prompts automatically in under 50ms - Single static Rust binary, zero cloud dependency, Python SDK - Works with Ollama (Robin's default local LLM backend) ## Robin-specific entity schema ideas - **Apps:** installed apps, equivalents found, ones the user likes/dislikes - **Commands:** commands the user has run, errors encountered, fixes applied - **Preferences:** DE/WM settings, theme, input devices, accessibility needs - **Migration state:** what has/hasn't been migrated from Windows/macOS - **Problems:** recurring issues and their resolutions ## Architecture ``` Robin frontend → Robin backend (FastAPI) → mnemo sidecar (Rust) ↓ SQLite knowledge graph (persists across sessions) ``` ## Why this matters for Robin's ND/adaptive user base ND users often have inconsistent session patterns — they may not follow up on a problem for days. mnemo's persistent graph means Robin can pick up exactly where it left off without requiring the user to re-explain context. This is a significant accessibility win. ## Linked - circuitforge-core: mnemo as cf-core memory module (shared implementation)

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference: Circuit-Forge/robin#17

No description provided.