@tpsdev-ai/langgraph-flair

LangGraph BaseStore adapter backed by Flair — durable agent memory with crypto-pinned per-agent identity, federated peer-to-peer sync, and cross-orchestrator portability.

Drop-in for LangGraph's InMemoryStore. The same memories your LangGraph agent writes are then visible to every other Flair-enabled harness:

• Claude Code / Cursor / Continue.dev / Codex (via @tpsdev-ai/flair-mcp)
• OpenClaw (via @tpsdev-ai/openclaw-flair)
• n8n (via @tpsdev-ai/n8n-nodes-flair)
• Hermes Agent (via hermes-flair)
• Pi agent (via @tpsdev-ai/pi-flair)

Install

npm install @tpsdev-ai/langgraph-flair @tpsdev-ai/flair-client
# Or, if you're already using LangGraph:
npm install @tpsdev-ai/langgraph-flair

Usage

import { FlairStore } from "@tpsdev-ai/langgraph-flair";
import { StateGraph } from "@langchain/langgraph";

const store = new FlairStore({ agentId: "my-agent" });

const graph = new StateGraph(...)
  .compile({ store });

// Or with createReactAgent:
import { createReactAgent } from "@langchain/langgraph/prebuilt";
const agent = createReactAgent({ llm, tools, store });

How LangGraph's namespace maps to Flair

LangGraph's BaseStore uses hierarchical namespaces (["users", "profiles"]) and string keys ("user123"). FlairStore maps each item to a Flair memory:

| LangGraph | Flair | |-----------|-------| | namespace: ["users", "profiles"] | tags: ["lg-ns:users/profiles"] | | key: "user123" | id suffix: lg:<agentId>:users/profiles:user123 | | value: { name: "Alice" } | content: '{"name":"Alice"}' | | search.query: "..." | semantic search via Flair's HNSW index | | search.filter: { age: { $gte: 18 } } | applied client-side after retrieval |

Authentication

FlairStore inherits from FlairClient. Three options:

1. Ed25519 keypair (preferred): set FLAIR_AGENT_ID and the client auto-resolves your key.
2. Explicit key path: new FlairStore({ agentId, keyPath: "/path/to/key.pem" })
3. Basic auth fallback: new FlairStore({ agentId, adminUser, adminPassword }) for standalone deployments.

const store = new FlairStore({
  agentId: "my-agent",
  url: "https://flair.example.com",  // or FLAIR_URL env var
  adminPassword: process.env.FLAIR_ADMIN_PASS,
});

What you get

• Persistence: memories survive process restarts and re-deploys.
• Federation: pair your local Flair to a hub; memories sync peer-to-peer.
• Cross-orchestrator: switch from LangGraph to OpenClaw to Claude Code without losing the agent's history.
• Identity: every memory is tied to a crypto-pinned agentId. No tenant-isolation slop.
• Open source: runs on your hardware. No SaaS lock-in.

Limitations (v1)

• LangGraph's IndexConfig (custom embedding model, per-field indexing) is ignored. Flair has its own embedding pipeline (nomic-embed-text-v1.5, 768-dim) and embeds the full content blob. If you need per-field embeddings, pre-extract and store as separate items.
• search.filter operators ($eq, $ne, $gt, $gte, $lt, $lte) are applied client-side after retrieving the namespace prefix. Tag-based pre-filtering (the namespace) keeps this bounded; high-fanout filters across many memories will be slower.
• Namespace-prefix matching uses the full joined-path tag (lg-ns:users/profiles). Items in ["users", "profiles", "u123"] are reachable via the ["users", "profiles"] prefix because the search routine post-filters parsed namespaces against the requested prefix — but searching by a single label anywhere in the namespace (e.g. "all items with profiles somewhere") isn't supported. LangGraph's BaseStore.search API doesn't expose this surface either, so there's no read path that would benefit; if a future LangGraph extension adds it we'll add a derived index then.
• listNamespaces returns namespaces seen in your stored memories (best-effort scan). Empty namespaces aren't enumerable.

License

Apache 2.0 — same as Flair core.