grapeot/context-infrastructure

What it is

Context Infrastructure is a reference implementation of a year‑long “context” system for an AI assistant. It shows how to organise data, let the model remember past interactions, and personalise its behaviour. The repo is not a ready‑to‑run product; it is meant to be cloned, examined, and adapted as a blueprint for your own AI‑driven workflow.

Core ideas

Layer	What it contains	How you use it
Presentation layer (read‑only)	rules/axioms/ (43 distilled decision axioms) and rules/skills/ (39 skill markdowns) – examples of the author’s accumulated knowledge.	Look at the structure and wording to inspire your own rules; do not copy them verbatim as they are personal to the author.
Reusable layer (editable)	Templates: rules/SOUL.md (AI identity), rules/USER.md (your preferences), rules/COMMUNICATION.md (style guide). Small scripts in periodic_jobs/ai_heartbeat/ that implement a three‑level memory system.	Fill the templates with your own data, configure the scripts, and run them (usually via cron) to give the assistant a personalized context.
Non‑reusable layer	The concrete content of the axioms and the experience behind each skill.	Use them as a model for how to capture and organise your own observations and reflections.

Quick start (≈5 min)

git clone https://github.com/grapeot/context-infrastructure
cd context-infrastructure
# Open the folder in Claude Code / OpenCode / Cursor (any IDE you like)

1. Open rules/USER.md and fill in your basic information – this is the step that immediately personalises the AI.
2. Follow the detailed steps in setup_guide.md to finish configuration (environment variables, cron jobs, etc.).
3. If you want to add extra capabilities, consult docs/SKILL_ECOSYSTEM.md, which lists optional public‑skill repos (web search, Google Docs, maps, email, etc.).

Repository layout (high‑level)

• Root files – routing table (AGENTS.md), setup guide, .env.example.
• docs/ – cron‑job guide and a catalogue of optional skill repos.
• rules/ – templates (SOUL.md, USER.md, COMMUNICATION.md), a workspace index, 43 decision axioms, and 39 skill markdowns.
• contexts/ – where the system stores its memory:
  • memory/OBSERVATIONS.md – the three‑level L1/L2 memory store.
  • daily_records/, survey_sessions/, thought_review/ – folders for raw logs, reports, and reflection notes.
• periodic_jobs/ai_heartbeat/ – code that runs daily (observer.py) and weekly (reflector.py) to update the memory store; includes design docs.
• tools/ – a report‑sharing helper (Tier 2).
• adhoc_jobs/ – place for one‑off tasks you might add.

How it works (conceptually)

1. Data collection – you (or other scripts) write observations into contexts/memory/OBSERVATIONS.md and other daily logs.
2. Periodic jobs – a cron‑triggered observer.py ingests new logs each day; a weekly reflector.py processes them into higher‑level summaries.
3. Rule files – the AI reads rules/SOUL.md (its persona) and rules/USER.md (your preferences) at the start of each session, shaping its responses.
4. Skill plugins – optional public skill repos can be added to extend functionality (search, calendar, payments, etc.). The core repo stays lightweight; capabilities are added on demand.

Intended use case

• Individuals who want an AI assistant that remembers their past interactions, decisions, and personal preferences.
• Developers looking for a concrete example of how to structure a “memory‑augmented” AI system, including data flow, periodic processing, and rule‑based personalization.
• Anyone interested in the design of long‑term context management for LLM‑driven agents.

License

MIT – you can freely copy, modify, and redistribute the code, respecting the license terms.