minigraf-examples

Examples, integrations, and cookbooks for Minigraf -- part of the Minigraf ecosystem.

This repository is a standalone Rust examples crate. Each scenario is runnable with Cargo and uses the published minigraf crate.

Prerequisites

• Rust 1.85 or newer for edition 2024 crates.
• Cargo with access to crates.io.
• minigraf = "1.1" from crates.io. Cargo currently resolves this to minigraf v1.1.1.

Install and verify dependencies:

cargo test

Scenarios

Agentic Memory

Stores user preferences and project context in Minigraf, queries them back for an agent response, then writes a correction while keeping transaction-time history.

Run:

cargo run --example agentic_memory

Expected output:

Agent memory: remembered Alice prefers concise technical answers.
Agent memory: retrieved Alice's current project, minigraf-examples.
Agent memory: wrote a correction with transaction history intact.

Offline-First Mobile

Stores local task changes while disconnected, queries pending changes for a future sync pass, then records a synced state while retaining earlier transaction state.

Run:

cargo run --example offline_first_mobile

Expected output:

Offline mobile: stored two local task changes while disconnected.
Offline mobile: selected the pending changes for later sync.
Offline mobile: marked the synced task without losing local history.

State Machine

Initializes an order in :awaiting-payment and stores legal transitions as Minigraf facts. A Datalog guard query checks whether an event is legal from the order's current state. A legal payment transition is accepted; an illegal ship event is rejected. The state update is an atomic write transaction containing a retract and a transact. A transaction-time replay with :as-of 1 shows the prior state.

Run:

cargo run --example state_machine

Expected output:

State machine: accepted payment by querying transition facts as the guard.
State machine: rejected shipping from awaiting-payment before the transition.
State machine: replayed transaction history to explain the prior state.

Audit Log

Records a policy approval, supersedes the owner, then queries both current state and an earlier transaction-time view.

Run:

cargo run --example audit_log

Expected output:

Audit log: recorded policy approval and superseding revision.
Audit log: queried the current policy owner.
Audit log: queried transaction-time history for the earlier owner.

LangChain Integrations

The LangChain examples use the published language bindings directly:

• Python: minigraf==1.1.1
• Node.js: minigraf@1.1.1

Python LangChain

Implements BaseChatMessageHistory from langchain-core with Minigraf-backed message storage.

Install prerequisites:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install -r integrations/langchain-python/requirements.txt

Run:

python integrations/langchain-python/minigraf_chat_history.py

Expected output:

Human: Remember that Minigraf stores agent memory.
AI: Got it. I will use Minigraf-backed chat history.

Ecosystem Crates

minigraf-algorithms

minigraf-algorithms/ is a standalone crate for graph algorithms that operate on Minigraf data. It lives outside Minigraf core so traversal helpers can evolve as opt-in ecosystem utilities without enlarging the embedded database API.

Run:

cargo test -p minigraf-algorithms

LangChain.js

Implements BaseChatMessageHistory from @langchain/core/chat_history with Minigraf-backed message storage.

Install prerequisites:

cd integrations/langchain-js
npm install

Run:

npm start

Expected output:

Human: Remember that Minigraf stores agent memory.
AI: Got it. I will use Minigraf-backed chat history.

GraphRAG Integration

The GraphRAG example uses ChromaDB for semantic retrieval and Minigraf for structured graph traversal. Entity UUIDs are the explicit bridge between the two stores.

• Python: minigraf==1.1.1, chromadb

Python GraphRAG

Populates a six-node concept graph in Minigraf, indexes entity descriptions in an in-memory ChromaDB collection, retrieves the closest entity by semantic similarity, then traverses its graph neighbours.

Install prerequisites:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install -r integrations/graphrag-python/requirements.txt

Note: ChromaDB downloads the all-MiniLM-L6-v2 embedding model (~80 MB) on first run.

Run:

python integrations/graphrag-python/graphrag_minigraf.py

Expected output:

Query: "storing time-varying relationships"
Match: temporal-data
Related: graph-db, knowledge-graph

LlamaIndex Integration

MinigrafGraphStore implements LlamaIndex's SimpleGraphStore with Minigraf as the backing store. Triplets map directly onto Minigraf's native entity-attribute-value model: subjects and predicates become Minigraf keywords (:myapp, :depends-on), objects are string values. The example shows tech-stack dependency evolution: a v1 graph is ingested, a dependency is upgraded, the current state is queried via get_rel_map, and Minigraf's :as-of <tx> temporal query recovers the pre-upgrade state from transaction history.

• Python: minigraf==1.1.1, llama-index-core

Python LlamaIndex

Subclasses SimpleGraphStore from llama_index.core.graph_stores with Minigraf-backed triplet storage and a raw Datalog pass-through in query().

Install prerequisites:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install -r integrations/llamaindex-python/requirements.txt

Run:

python integrations/llamaindex-python/minigraf_graph_store.py

Expected output:

myapp depends-on: pydantic==2.0, requests==2.28
requests depends-on: urllib3==1.26
myapp at tx 3 depended-on: pydantic==1.10, requests==2.28

Temporal model: Each upsert_triplet call is one Minigraf transaction. After three ingestion calls, SNAPSHOT_TX = 3. Retracting and re-asserting a dependency advances the transaction counter. (query [:find ?o :as-of 3 :where [:myapp :depends-on ?o]]) returns the dependency set as it existed after transaction 3 — before the upgrade.