backpack-ontology
A persistent knowledge graph engine for Claude Code. Gives your AI agent structured, searchable memory that persists across sessions — store
Open source Repository Open in the app JSON README (API)
About
A persistent knowledge graph engine for Claude Code. Gives your AI agent structured, searchable memory that persists across sessions — stored as typed graphs with nodes and edges. Features progressive discovery so the context window stays clean: Claude browses layer by layer, only pulling in what it needs. Supports freeform schemas, full-text search, graph traversal, and bulk operations. Data is stored as human-readable JSON following the XDG specification.
Details
- Kind
- Plugins
- Topic
- AI, RAG & memory
- Publisher
- noahirzinger
- Origin
- marketplace
- Category
- ferramentas
- Last push
- 2026-04-16T19:36:29Z
- Repository state
- ativo
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
noahirzinger/backpack-ontology-plugin/backpack-ontology
README
# Backpack Plugin **Carry your knowledge forward.** Backpack lets Claude remember what matters: your clients, your processes, your decisions. Knowledge that travels with you. ## What it does Backpack is a persistent knowledge base that Claude carries across conversations. Inside it are learning graphs, each about a different topic. There are no enforced schemas. Claude decides what structure fits the domain. ## Install ### Claude Cowork (web) Tell Claude: > "Install the backpack plugin from github.com/NoahIrzinger/backpack-ontology-plugin" Claude will set it up for you. After that, just start talking. ### Claude Code (terminal) **1. Register the marketplace and install the plugin:** ``` /plugin marketplace add NoahIrzinger/backpack-ontology-plugin /plugin install backpack-ontology@NoahIrzinger-backpack-ontology-plugin ``` The first command registers this repo as a marketplace. The second installs the plugin from it. **2. Enable auto-update (recommended):** Run `/plugin` to open the plugin manager, go to the **Marketplaces** tab, select `NoahIrzinger-backpack-ontology-plugin`, and toggle **Enable auto-update**. This is a one-time opt-in — third-party marketplaces default to auto-update off, which means you'd otherwise need to manually reinstall every time a new version ships. With auto-update on, Claude Code checks for new versions at startup and rolls them forward automatically. **3. Restart Claude Code** (or run `/reload-plugins`) and you're ready. For local development against a cloned repo, start Claude Code with: ``` claude --plugin-dir ./backpack-ontology-plugin ``` ### Connect to the cloud (free) Sign up for a free account at [app.backpackontology.com](https://app.backpackontology.com) to sync across devices and share with your team: ``` claude mcp add backpack-app -s user --transport sse https://app.backpackontology.com/mcp/sse ``` ## What to say to Claude No commands to learn. Just talk naturally. ### Remember something > "Remember that Acme Corp is on the Enterprise tier, main contact is Sarah Chen" > "Start a learning graph for our hiring process" > "Add the new vendor agreement to backpack" ### Mine a topic autonomously > "Mine the gut microbiome into a learning graph" > "Build a learning graph about Spanish wines, 30 nodes max" > "Research the history of postgres into the backpack" Claude will run an iteration loop: find sources, extract entities and relationships, validate against the existing graph, and commit. It stops when it hits the iteration cap, the node target, or when it stops finding new entities. Every node carries the source it came from, so the graph stays traceable. ### Extract a graph into documents > "Extract the spanish-wines graph" > "Generate docs from my research graph" After mining, Claude generates KB documents from the graph: overview, taxonomy, key entities, relationship patterns, findings, and source provenance. ### Refine signals > "Refine my backpack" > "What signals are in the gut-microbiome graph?" Runs structural detectors across all graphs to surface what matters: gaps, imbalances, disconnected clusters, cross-graph inconsistencies. ### Full pipeline > "Mine spanish wines, extract the core KBs, and refine the signals" > "Full pipeline on the gut-microbiome" Chains all three steps: mine → extract → refine. ### Find something > "What's in my backpack about Acme Corp?" > "Search backpack for anything related to compliance" ### See the big picture > "Show me my learning graph" > "What's in my backpack?" Claude will open a graph visualizer so you can explore your knowledge visually. ### Move to the cloud > "Sync my backpack to the cloud" Sign up for a free account at [app.backpackontology.com](https://app.backpackontology.com) to access your knowledge from any device and share with your team. ## What's included | Component | What it does | |---|---| | **MCP Server** | Tools for storing, searching, traversing, auditing, and normalizing learning graphs. Auto-migrates pre-0.3.0 graphs on first start. | | **`backpack-guide` skill** | Teaches Claude best practices for organizing and querying your backpack | | **`backpack-mine` skill** | Autonomous mining loop — point Claude at a topic and grow a graph from web sources | | **`backpack-extract` skill** | Generate KB documents (overview, taxonomy, key entities, findings) from a mined graph | | **`backpack-refine` skill** | Run signal detection and enrich findings with contextual descriptions | | **Visualization** | Claude launches a web-based graph explorer when you ask to see your data | ## Reference ### CLI commands | Command | What it does | |---|---| | `npx backpack-viewer@latest` | Open the graph visualizer (http://localhost:5173). Always include `@latest` — `npx backpack-viewer` without the version suffix reuses a cached older version. | | `npx -p backpack-ontology@latest backpack-sync` | Upload local learning graphs to Backpack App | | `npx -p backpack-ontology@latest backpack-init` | Remove any leftover Backpack hooks from `.claude/settings.json` (the MCP server also runs this cleanup automatically on startup) | ### Data storage Local learning graphs are stored at `~/.local/share/backpack/graphs/<name>/`. Each graph has an append-only event log (`branches/<branch>/events.jsonl`) and a materialized snapshot cache (`branches/<branch>/snapshot.json`). Graphs from older versions are migrated automatically on first start. ## Privacy & Telemetry See the [Privacy Policy](https://github.com/noahirzinger/backpack-ontology/blob/main/PRIVACY.md). Anonymous usage telemetry is collected by default (tool counts, session duration, never content). Opt out with `DO_NOT_TRACK=1`. ## License Apache-2.0