🧠 Mimir – The Well of Wisdom

Unified data orchestration system for personal and project-level intelligence. Core philosophy: “ingest once, know forever” — data may disappear, but insight persists.

Overview

Named after the Norse well of wisdom, emphasizing knowledge retention
First sources: GitHub (commits, PRs, comments) and Strava (activities)
Long-term vision: self-evolving knowledge base that can reason over its data
Open-source-ready with transparent manifests, schemas, and code

Architecture

Execution Model

User-land daemon: mimir daemon
Manual jobs: mimir run-once foo
Actions are declarative manifests, executable independently or via scheduler
Event system (events table) records ingestions, transforms, detections, reports
Idempotency is core — every record uniquely traceable

Storage Tiers

Hot: Postgres (active data)
Warm: Parquet/S3 (archived data)
Cold: Glacier (long-term storage)
Schema validation via JSON Schema (strict in prod, tolerant in dev)
Versioning by dataset (vX.Y.Z) with full lineage tracking

Data Flow

Ingestion: mostly pull, sometimes push (webhooks, filters, summaries)
Data can come from other Mimir outputs → recursive aggregation
Transforms produce versioned snapshots (no overwrite policy)
Mutability: patch when confident, reprocess when uncertain

Outputs

Reports: HTML + PNG by default (Markdown/PDF later)
Dashboards: Real-time monitoring views
Detectors: Anomaly detection and alerts
Telegram Bot: Mobile interface for queries and notifications

Telegram Bot Features

Personal Telegram bot interface for on-the-go Mimir access
Query data naturally: “show me GitHub activity this week”
Subscribe to reports: daily summaries, anomaly alerts
Generate visualizations on demand with inline PNG responses
Voice message transcription for hands-free data queries
Proactive insights: “Your Strava pace improved 5% this month”
Integration with mnemos-core-features|Mnemos: “save this insight as a fact”

Open Questions

Data Model

What’s the minimal shared schema? (source, entity, ts, tags, payload)?
How to handle schema evolution across versions?
Should we use unified or source-specific schemas?

Architecture

How to define action manifests (YAML schema)?
Should daemon watch files, schedules, or both?
Where to store patch history for mutations?
How to expose lineage graph (e.g., /lineage or visual DAG)?

Features

Which alerting rules for v0.1?
Should bot embed LLM features from day one?
How much automation vs manual control initially?
Should reports include diffs by default?

Integration with Mnemos

Mimir and p/moul/mnemos|mnemos form a data-to-wisdom pipeline:

Data Flow

Mimir ingests and orchestrates data → p/moul/mnemos|mnemos structures knowledge from it
Raw events → insights → structured knowledge → wisdom
Mimir’s event system could feed Mnemos for automatic note generation
Mnemos questions could drive new Mimir data collection actions

Synergies

Mimir reports become Mnemos facts
Mimir anomalies become Mnemos questions
Both tools share philosophy of knowledge persistence and evolution
Together: data orchestration + knowledge management = intelligence system

This integration is part of the kb/tool-trinity|moul tool trinity vision.

mimir