Recall vs LangGraph checkpointers
A LangGraph checkpointer saves the whole graph state after every super-step so a thread can continue, time-travel or wait on an interrupt. Recall records what the agent knows and where it is, so a new process, a new image or a different framework can pick up from a briefing. They are not rivals: recall-langgraph ships RecallSaver, a checkpointer that passes LangGraph's conformance suite, and RecallStore, next to RecallResume.
Recall and LangGraph checkpointers and stores
| Recall | LangGraph checkpointers and stores | |
|---|---|---|
| Storage ownership | Your own database: polign_db in your S3, GCS or Azure bucket, or PostgreSQL, MongoDB or Qdrant you already run. No vendor control plane. | Your own database through a checkpointer package: PostgreSQL, SQLite, MongoDB, Redis, Cosmos DB, or in memory. Stores (BaseStore) likewise. LangSmith Deployment runs its own. |
| Typed records | Yes. Each fact is a subject, a predicate from a registry (or coined on first use) and a typed value; text is extracted into those by a model you name. | A checkpoint is the full graph state (every channel's value) at each super-step. A store holds JSON items in tuple namespaces with optional embedding search; LangMem adds memory tools and a background extractor over it. |
| Corrections and supersession | By rule at read time: a single-valued predicate's newer statement replaces the older one; a text statement can name what it makes untrue. Reads carry what was replaced. | Not a concept of a checkpointer. In a store a put overwrites the item under its key; what the old value was is gone unless you keep it yourself. |
| Resume after restart | Yes. Working state, turns, outputs, pointers and in-flight tool calls per agent id, fenced by a lease; a new process resumes from a briefing. | Yes, within a thread: the graph restarts from the last durable checkpoint at a super-step boundary. A node that was running is re-run from its start, so side effects must be idempotent; nothing settles a tool call the crash interrupted. |
| Shared across agents | Yes. One database, keys per namespace through recall serve, Kubernetes workload identity, packages for LangGraph, OpenAI Agents SDK, Pydantic AI, LiveKit, Vapi and MCP hosts. | Within LangGraph, yes: one store across threads and graphs. Other frameworks need their own path to the same database. |
| Audit and history | Append-only records with time and source; history, as_of reads, diff, and audit bundles replayable offline. | Every checkpoint is kept, so a thread's state history is browsable (get_state_history). Stores keep no history. TTL can expire both. |
| Self-hosted path | pip install polign-recall runs a local database; recall serve or the Helm chart for a shared one. Apache-2.0. | pip install langgraph-checkpoint-postgres against your PostgreSQL; MIT. |
| Pricing | Free to self-host, no limits. Paid: design partner seat (60 days free, then $12k for 12 months) or a Production Agreement at $24k a year, flat. | Open source is free. LangSmith Deployment is priced per seat and per node, with managed persistence included. |
When to pick the other tool
Pick a plain checkpointer, PostgresSaver for production, when your agent is one graph on one thread and what you need is to continue it, interrupt it for a human, or replay from an earlier checkpoint; that is what checkpoints are for, and they are exact. Add LangMem on a store when free-text memories across threads are enough. Reach for Recall when the memory must be correctable by rule and readable from outside LangGraph; when the agent has to come back on a different image or model without the old state; when tool calls in flight at the crash must be settled rather than re-run; or when you want the checkpointer, the store and the memory in one database you own. On the resume gate, agents resumed from a briefing on a fresh thread finished 20 of 20 tasks at the same token cost as uninterrupted runs.
Where to go next
- recall-langgraph: RecallResume, RecallSaver and RecallStore
- Resume an agent after a crash or pod restart
- Durable agent actions: tool calls a crash interrupted
Sources, read on November 3, 2026: LangGraph checkpointers (checkpoints, durability modes, pending writes), stores, durable execution, langgraph-checkpoint-conformance, LangMem; langgraph-checkpoint 4.3.1.
Recall is typed, durable memory and state for AI agents. It stores corrections, decisions and task state as append-only records in your own object store (S3 or GCS), so an agent can be corrected without retraining and resumed after a crash, restart or pod replacement without snapshotting the process.
Try it
One pip install brings the client, the recall binary and the database. The snippet on
this page runs as written.