ai-agent-learnings

KO | EN

gitlite — search

#javascript #python #hacktoberfest #typescript #rust #deep-learning #linux #macos #nodejs #vue #security #ios #openai #open-source

ai-agent-learnings

(★ 9)

Patterns and architectures for AI agent knowledge management systems — battle-tested in production

ai-agent-learnings Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# AI Agent Learnings

**This content has moved to [jibot.md/learnings](https://jibot.md/learnings/index.md).**

The documents below are archived. For the latest versions, visit jibot.md.

---

Patterns for building AI-operated knowledge management systems, extracted from
a production system (~5,700 indexed files, multiple agents, multiple machines,
daily use since 2024).

*Last updated: 2026-03-26*

## Documents

| Document | What It Covers |
|----------|---------------|
| [architecture.md](architecture.md) | The complete system: three-tier pipeline, routing, frontmatter contracts, containment, reweave, Seven Gates audit, multi-agent boundaries, heartbeat triage, search infrastructure, LLM routing, implementation checklist |
| [knowledge-systems-audit.md](knowledge-systems-audit.md) | Full system audit: component inventory, retrieval systems, sync topology, integration matrix, data flow gaps, Python module status, design doc registry |
| [tiered-access-design.md](tiered-access-design.md) | Multi-agent security model: four access tiers, signal group architecture, OpenClaw routing, document airgap strategy, migration phases |
| [cross-pollination.md](cross-pollination.md) | What we learned from four external projects (Ars Contexta, Mechanics of Magic, Ethoswarm, Colin Raney & Fred) and what we adapted |
| [ethoswarm-switchboard-bridge.md](ethoswarm-switchboard-bridge.md) | Integration spec for connecting Ethoswarm persistent agents with our knowledge pipeline (RETIRED 2026-07-12 — Ethoswarm deprecated, integration dismantled) |
| [collaborative-knowledge-domains.md](collaborative-knowledge-domains.md) | Architecture for multi-human, multi-agent collaborative knowledge domains using file-based intake/curation/atlas pipeline synced with Syncthing -- evaluated against 7 alternative architectures |
| [domain-operations-guide.md](domain-operations-guide.md) | Practical companion: full folder structure, three operating modes (automated curation, interactive tending, publication), versioning system, PDF generation, maintainer agent pipeline, setup checklist |
| [domain-context-template.md](domain-context-template.md) | Copy-paste template for CONTEXT.md -- the session orientation file that makes interactive AI tending sessions productive from the first prompt |
| [multi-interface-knowledge-layer.md](multi-interface-knowledge-layer.md) | Projecting a personal vault into a team-facing document platform (Onyx) with multi-interface adapters (chat bot, Slack, email), cross-channel deduplication, canonical identity resolution, and policy engine separation |
| [onyx-rag-deployment.md](onyx-rag-deployment.md) | Deploying Onyx (open-source RAG platform) as a team-facing knowledge interface: vault projection pipeline, zero-fork customization strategy, custom AI agents, multi-channel governance, infrastructure and cost, lessons learned |

## If You Only Read One Thing

Read [architecture.md](architecture.md). It contains everything: the pipeline,
the principles, the schemas, the audit framework, and the implementation
checklist. The other documents are supplementary.

## System Status (2026-03-26)

The system is **OPERATIONAL** with nine major infrastructure components deployed:

| Component | Status | What It Does |
|-----------|--------|-------------|
| **Heartbeat** | Live | Automated 15-min collection and triage cycle |
| **Ethoswarm Curator Mind** | Retired 2026-07-12 | Was always-on knowledge intake via Telegram; Ethoswarm deprecated, integration dismantled |
| **QMD Search** | Live (~4,300 files) | Three-mode search: keyword, vector, deep |
| **iblai-router** | Live (since Mar 4) | Intelligent LLM routing, 80% cost savings |
| **Amplifier Bundle System** | Live | 18+ tool modules, multi-provider routing, recipe workflows |
| **Talk Pipeline** | Live | research, outline, produce, slides (Claude + Gamma) |
| **People Pipeline** | Live | harvest from daily notes, Nyne enrichment, network review |
| **Meeting Pipeline** | Live | muesli sync, translation, extraction, daily note injection |
| **Onyx (Team Knowledge)** | Live | RAG + chat interface for partner org, vault projection pipeline, multi-channel governance |

## The System

Built on:
- **Obsidian** (markdown + wikilinks) as the knowledge substrate
- **Syncthing** (peer-to-peer) for multi-machine sync with containment boundaries
- **Amplifier** (github.com/microsoft/amplifier) as the multi-agent framework
- **GTD** (Getting Things Done) for connecting knowledge to action
- **QMD** for three-mode search (keyword, vector, deep) over indexed collections
- **sprites.dev** for cloud microVM infrastructure (knowledge-intake sprite)

## Who This Is For

- **AI agents** building or advising on knowledge systems (documents are agent-ingestible)
- **Developers** building multi-agent systems that need structured knowledge
- **Anyone** running a personal knowledge management system at scale

ai-agent-learnings

ai-agent-learnings Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

ai-agent-learnings

ai-agent-learnings Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README