sibyl-memory-langgraph
A LangGraph BaseStore backed by
Sibyl Memory — durable, long-term, cross-thread memory
for your agents on SQLite + FTS5. No vector database, no embeddings.
Install
pip install sibyl-memory-langgraph
Part of the Sibyl Memory plugin family. See the CHANGELOG for release notes.
from sibyl_memory_langgraph import SibylStore
from langgraph.graph import StateGraph
store = SibylStore() # ~/.sibyl-memory/memory.db; needs `sibyl init` once
graph = StateGraph(State, store=store)
Direct use:
store.put(("memories", "u1"), "fact1", {"text": "prefers dark mode"})
item = store.get(("memories", "u1"), "fact1")
hits = store.search(("memories",), query="dark mode") # lexical, subtree
names = store.list_namespaces(prefix=("memories",))
Mapping
| LangGraph | Sibyl Memory |
|---|---|
namespace tuple |
category ("/".join(namespace)) |
key |
entity name |
value dict |
entity body (JSON) |
Scope
- Long-term Store only (not a checkpointer).
searchis lexical FTS5, not vector similarity.PutOp.indexandPutOp.ttlare accepted and ignored (no embedding index, no TTL).- Namespace elements must be non-empty and contain no
/or...
Identity
SibylStore() with no explicit client or tenant_id binds to the activated
account: it reads ~/.sibyl-memory/credentials.json (written by sibyl init,
looked up next to the DB file) and resolves the tenant via the canonical ladder
credentials.tenant_id -> credentials.account_id -> DEFAULT_TENANT
DEFAULT_TENANT is a constant for tests and for data written before accounts
existed; that data moves into the account tenant the first time a signed-in
client uses the store (client 0.9.0, local only), or with sibyl memory adopt. The credentials file is symlink-guarded: a symlinked credentials.json
is treated as absent rather than followed. Pass tenant_id="..." to override, or
client=my_memory_client to use that client's tenant as-is.
Account required (0.3.0)
Every store operation needs an activated account (sibyl init) holding a valid,
server-signed entitlement for this computer. Without one, SibylStore() still
constructs, but put, get, search, delete, list_namespaces (and
batch / abatch) raise sibyl_memory_client.EntitlementError. A framework
store propagates it, and LangGraph surfaces it to the graph. e.code says why
(no_account, signed_out, expired_offline, ...) and the message names the
one action (usually sibyl init). tenant_id= (including DEFAULT_TENANT)
chooses the tenant, never the access. The gate reads the credentials.json
next to the DB file when there is one, else ~/.sibyl-memory/credentials.json.
For an agent driving the graph, the exception carries the handoff (sibyl-memory-client
0.9.0): e.next is the exact command to run next (sibyl init, sibyl refresh,
sibyl init --force, ...; or done), and e.human_action is the sentence to relay to
the person, or None when nobody has to act. An uncaught traceback shows the same two
lines at the end of str(e) (ask your human: ... when a person must act, then
next: ..., always last); e.summary is the message without them.
from sibyl_memory_client import EntitlementError
try:
store.put(("notes",), "k", {"v": 1})
except EntitlementError as e:
if e.human_action:
print("ask your human:", e.human_action)
print("next:", e.next)
Embedding or testing: pass your own client, for example one built with the client's test gate:
from sibyl_memory_client import MemoryClient
from sibyl_memory_client.testing import NullEntitlementGate
store = SibylStore(client=MemoryClient.local("/tmp/t.db", entitlement_gate=NullEntitlementGate()))
Local-first & telemetry
Memory reads and writes are fully local: a SQLite database in
~/.sibyl-memory/, and memory content never leaves the machine. This adapter
inherits the posture of the underlying sibyl-memory-client, which talks to
api.sibyllabs.org for these things only:
- activation (
sibyl init); - a daily account check-in that renews this computer's access (account id, this computer's sign-in token, plugin version, OS family; no memory content);
- the size check when a free account nears its cap (account id, sign-in token, the database's byte size and the proposed delta, and this computer's signed access token: account id, tier, cap, device hash, account flags; no personal data);
- a debounced usage heartbeat: an aggregate operation count only, never
memory content, query text or entity names, sent with this computer's
signed access token (the same fields as above; no personal data). Opt out
with
SIBYL_MEMORY_TELEMETRY=0.
Your local data is always readable with sibyl memory list|search|recall,
with or without an account.
MIT. Built by Sibyl Labs, LLC.
Save and debloat
The plugin's save and debloat procedures belong to agent sessions (the MCP server, the Hermes
adapter, the Claude Code hooks), not to a framework store, so this package has no lifecycle of
its own. A LangGraph agent that also uses the MCP server gets the save and debloat prompts
there. The Pro automatic save (by active time) and the skill proposals ride the MCP server and
the Hermes adapter too; SibylStore never asks for a save.
Uninstalling the CLI: run sibyl skills uninstall before pip uninstall sibyl-memory-cli; it
removes the Claude Code hooks that call the sibyl command (they exit quietly once it is gone,
but stay in ~/.claude/settings.json until removed).
Metadata
Release files for sibyl-memory-langgraph 0.3.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| sibyl_memory_langgraph-0.3.0.tar.gz | 70.1 kB | Without provenance | Details for sibyl_memory_langgraph-0.3.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| sibyl_memory_langgraph-0.3.0-py3-none-any.whl | Python 3 | none | any | Details for sibyl_memory_langgraph-0.3.0-py3-none-any.whl |
Total release size: 85.6 kB
Release files / sibyl_memory_langgraph-0.3.0.tar.gz
| Download URL | sibyl_memory_langgraph-0.3.0.tar.gz |
|---|---|
| Size | 70.1 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
8aaa5bb3648c6b87f87b06be8ac306665c556888b0cc55aa79e72457fd263fce
|
|
BLAKE2b-256 checksum How to use checksums |
5fb97ee13e06d49d0801b436a4d1049ad944ca120a4fcf923eb8014c00d45d3c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.3
|
Release files / sibyl_memory_langgraph-0.3.0-py3-none-any.whl
| Download URL | sibyl_memory_langgraph-0.3.0-py3-none-any.whl |
|---|---|
| Size | 15.5 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
77a954cf6ef91d439253faa858f3ba0f3b7a1288140895f14c64594650ae5183
|
|
BLAKE2b-256 checksum How to use checksums |
4ae2aa809862b041e1fecce890fa500ce49937808683f69cca0b8b5940c47f88
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.3
|
Security for sibyl-memory-langgraph 0.3.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.