TL;DR
Relate is SQLite for property graphs: a Rust, in-process, in-memory graph engine that speaks Cypher 25, persists to one .rl file or a directory, and can be served unchanged as Bolt, HTTP and MCP. For agent memory it removes the server you would otherwise run (Neo4j, a vector DB) and adds vector, full-text and spatial search on the same graph.
What it buys: zero infrastructure, one file per agent or per project, Neo4j driver and tooling compatibility, and an agent-facing surface designed to be hard to misuse (in-band row caps, structured retryable errors, a read tool that provably cannot write).
What it costs: the whole graph lives in RAM, one process owns the store, there is no general write-skew detection, and it publishes no benchmarks. It is an experimental Neo4j Labs project, v1.0 released 2026-11-11, one week after the repo was created.
Verdict: a strong default for a single agent's (or single developer's) local graph memory and for agent-built knowledge graphs of code and files. Not a shared, multi-writer memory service.
Source
The post is a five-sentence note by André Lindenberg (posted 2026-11-12): agent memory does not need a server; a graph can live in one file the agent uses, and the same store can run as an MCP server; the cost is that the graph must fit in RAM and one process owns it. The repo link, relatedb/relate, is in the author's first comment, which calls it a store for "memory or transactional filesystem".
The repo is a Neo4j Labs project under Apache-2.0, mostly written by Michael Hunger with Oskar Hane and others. It accepts issues, not pull requests. It carries 500+ numbered architecture decision records and SSH-signed releases.
Architecture
One engine, embedded in the caller's process, with many front doors. The live graph is held in RAM in a typed CSR (compressed sparse row) layout for traversal. Durability is a fsync'd write-ahead log plus one snapshot codec (Parquet shards or a readable JSON file). Commits are optimistic: conflicts are detected on element IDs and on unique/key constraint values at a commit-time serialization point.
| Store | What it is | Durability |
|---|---|---|
:memory: | RAM only | gone at exit |
| Directory (default) | WAL + Parquet or JSON snapshot | full (ack after WAL barrier), everysecond (can lose ~1 s), or snapshot (loses everything since the last checkpoint) |
Single .rl file | WAL, snapshot and catalog in one host file | same three modes |
| Surface | How it runs |
|---|---|
| CLI and shell | relate binary; --json, --jsonl, or an envelope mode with a truncated flag |
| Rust, Python, Node | in-process; Python via PyO3 with zero-copy Arrow out to pandas or Polars |
| Browser | WASM, persisted to OPFS |
| Java, Go, Swift | wrappers over a C ABI |
relate serve | one process exposing Bolt (Neo4j drivers and browser work), HTTP Query API v2, and MCP over streamable HTTP; binds to localhost with a generated token by default |
The query language is Cypher 25: quantified path patterns, path selectors, schema DDL, 80+ functions, EXPLAIN/PROFILE and LOAD CSV. On top of the graph it adds vector properties with a SEARCH clause (top-K with pre-filtering), hybrid vector plus full-text search, R-tree spatial indexes, and pull/push of subgraphs to and from Neo4j or Aura. Conformance is measured by running the openCypher TCK and Neo4j's own Cypher suites on every ready PR, reported as coverage rather than enforced as a merge gate.
Agent-native design choices
The most transferable part of Relate is not the engine but how it treats an agent as the primary client. Each choice closes a specific way agents misread tool output.
Truncation is said in-band. Under a detected agent harness (or
--agent-mode), a statement prints at most 100 rows and says so in the output, so a capped result never reads as complete.--max-rowslifts it.Errors are data. In machine modes a failure prints
{"error":{code, exit_code, kind, message, retryable}}, and exit codes separate usage (2), rejected statements (6), write conflict or locked store (5) and other failures (1). The agent can branch on whether a retry can help instead of guessing from text.Read is provably read-only. The MCP surface is two tools,
read_cypherandwrite_cypher. Each statement is classified with EXPLAIN before it runs, soread_cypherrejects any write or schema change and a host can auto-approve it. In read-only modewrite_cypheris not offered at all.The skill ships with the binary.
relate skill installwrites a versioned SKILL.md bundle (CLI flags, Cypher dialect, vector and spatial references, known gotchas) into every detected agent (Claude Code, Cursor, Codex, Gemini CLI, Copilot and others). It never overwrites a hand-written skill directory, andrelate upgradekeeps it current.Extensions get only Cypher.
relate run <name>packages talk to the host over MCP stdio, and their only power is read/write Cypher. Shipped examples turn a directory tree (fs-to-graph, with full-text over file contents) or a git history (git-to-graph) into a graph.Typos get a suggestion. An unknown flag answers with the nearest real one, which saves an agent a round trip.
Tradeoffs and limits
The project's own limitations page is unusually candid. The ones that matter for agent memory:
| Limit | Consequence for agent memory |
|---|---|
| Whole graph in RAM; no paging, buffer pool or spill | Memory size is bounded by the host. An optional logical cap fails growing writes with Error::Capacity rather than swapping. |
| One process owns the store | Several agents sharing memory must go through one relate serve process, which becomes the single point of failure and contention. |
| Optimistic commits check element IDs and unique/key values only | Two agents that read the same facts and write different conclusions can both commit (write skew). Invariants must be expressed as constraints. |
| One graph per statement; single-level transactions | No cross-project memory joins and no nested save points. |
| Fixed query size, nesting and traversal-depth limits | Deep multi-hop recalls must be bounded or split. |
| No published benchmarks | Latency and memory per million edges are unknown; measure before adopting. |
| Anonymous CLI telemetry on by default | Turn it off (relate telemetry off or DO_NOT_TRACK=1) for private memory. |
| Experimental Labs project, one week old at v1.0 | API and on-disk format churn is plausible; Neo4j does not support it as a product. |
Failure modes
A graph store fixes the shape of memory, not its quality. The ways a Relate-backed memory goes wrong are mostly upstream of the engine.
Schema drift by the agent. An agent writing Cypher freely invents labels and relationship types (
:Fact,:Memory,:Note). Without schema DDL and constraints, recall queries silently miss half the graph.Stale facts stay authoritative. Nothing in the engine ages, versions or retracts a fact. Without a provenance edge (source, run, timestamp) and a garbage-collection pass, the graph is the memory-without-reconciliation failure R002 describes.
Vector recall without structure. Hybrid
SEARCHmakes it tempting to treat the graph as a vector store with extras. The value of the graph is multi-hop recall; if every read is top-K similarity, a flat vector index is simpler.Lost writes under relaxed durability.
everysecondcan drop about a second of acknowledged writes on a crash, andsnapshoteverything since the last checkpoint. An agent that wrote a memory and moved on will not notice.The serve process becomes shared infra anyway. Once two agents or a dashboard need the same memory,
relate serveis a server you run, with one process and no replication. The "no server" claim holds only for one owner.Auto-approved writes.
read_cypheris safe to auto-approve;write_cypheris not. A host that approves both lets a prompt-injected agent rewrite or delete its own memory (MATCH (n) DETACH DELETE n).
When to use it
Use Relate when one agent or one developer owns the memory, the graph fits comfortably in RAM, and recall is genuinely relational (which file calls which, which decision superseded which, who said what about whom). Prefer Markdown files plus ripgrep when memory is small and mostly prose, and a server database when several writers share memory or it must outgrow one machine.
Review checklist
Is recall multi-hop? If every query is "top-K similar notes", a vector index or plain files are simpler.
Is there a declared schema (labels, relationship types, unique/key constraints) that the agent's skill tells it to use?
Does every fact carry provenance (source, run, time) and is there a reconciliation or expiry pass?
Is durability
fullfor memory the agent will not re-derive?Is only
read_cypherauto-approved, with writes gated or in a separate read-only mount for untrusted runs?Is the expected graph size measured against host RAM, with the logical capacity cap set?
Is telemetry off for private stores?
Is there an export path (Parquet/JSON snapshot, or push to Neo4j) in case the Labs project stalls?
Related entries
| Entry | Link to Relate |
|---|---|
| R002 The 5 layers of agent memory | Relate is a store for the semantic and episodic layers; R002's warning that the missing step is reconciliation applies unchanged. |
| R028 auto-ontology-mcp | Same author, same pattern: a local, file-backed store behind MCP. R028 types memory in SQLite with validation; Relate makes it a graph with a read tool that cannot write. |
| R040 WikiTrace | WikiTrace's insight-to-run links are a natural graph; Relate could hold them, but the evidence-keeping discipline is the point, not the store. |
| R005 Second Brain OS | R005 keeps a linked-prose wiki; Relate is the typed, queryable version of those links. |
| R037 Floorplan | fs-to-graph and git-to-graph build the same kind of codebase map Floorplan draws, as a queryable graph. |
| R020 The model proposes, the harness enforces | Classifying each statement with EXPLAIN before running it is a harness-side check in R020's sense. |
| R054 Harness engineering | Counterweight: real coding harnesses use files and ripgrep, not databases, for memory. Relate must beat that baseline to earn a place. |
Sources
André Lindenberg's LinkedIn post, 2026-11-12
relatedb/relate on GitHub: README and docs/limitations.md
relatedb.ai, the project's documentation site