anansi-memory
Persistent memory for AI apps, backed by a bi-temporal knowledge graph — every fact carries when it was true and when you learned it. Give any LLM app long-term memory in two API calls. MIT-licensed and self-hostable.
pip install anansi-memory
Usage
from anansi_memory import AnansiMemory
memory = AnansiMemory(api_key="ans_...")
# Store a conversation turn
memory.ingest(
user_id="user_123",
content="User is building a voice agent. Prefers TypeScript. Team of 4.",
source_type="conversation",
)
# Before your next LLM call — inject synthesized context into the system prompt
ctx = memory.context(user_id="user_123", q="what is the user building?")
system_prompt = f"You are a helpful assistant.\n\n{ctx.format_for_prompt()}"
context() returns a synthesized profile — static facts and dynamic context that drop straight into a prompt, plus relevant chunks when you pass q. No chunks to dedupe, rank, or trim yourself.
API
AnansiMemory(api_key, base_url=None)
| Param | Type | Description |
|---|---|---|
api_key |
str |
Your API key (ans_...) |
base_url |
str |
Override API base URL (default: https://anansimemory.com) |
memory.ingest(user_id, content, source_type=None, source_id=None, metadata=None, embedding=None, session_id=None, agent_id=None)
Store content in a user's memory. Returns IngestResult(id, queued=True).
| Param | Type | Required | Description |
|---|---|---|---|
user_id |
str |
✓ | Your internal user ID |
content |
str |
✓ | Text to remember, max 100 KB |
source_type |
str |
"conversation", "document", "note", "meeting", "custom" |
|
source_id |
str |
Idempotency key — re-ingesting the same ID is a no-op | |
metadata |
dict |
title, author, timestamp, any custom fields |
memory.context(user_id, q=None, as_of=None, as_of_knowledge=None)
Retrieve synthesized memory for a user. Returns ContextResult(static, dynamic, relevant, ...). Pass as_of / as_of_knowledge for a bi-temporal point-in-time view (Pro+).
Also available
results = memory.search(user_id="user_123", query="dark mode", search_mode="hybrid")
chunks = memory.list_memories("user_123", source_type="conversation")
entities = memory.list_entities("user_123", as_of="2026-05-01", as_of_knowledge="2026-05-01")
memory.delete_user("user_123") # GDPR hard-delete
# LangChain-compatible retriever (no langchain dependency required)
from anansi_memory.langchain import AnansiRetriever
retriever = AnansiRetriever(api_key="ans_...", user_id="user_123")
docs = retriever.get_relevant_documents("what stack does the user prefer?")
list_entities accepts as_of (the graph as it was valid at an instant) and as_of_knowledge (the graph as we knew it at an instant). The entity graph is a Pro+ feature.
Handling API Responses
Error Handling
The Anansi SDK raises an AnansiError for non-successful API responses. You can use the status_code attribute to handle different types of errors programmatically.
from anansi_memory import AnansiMemory, AnansiError
import time
try:
memory.ingest(user_id=user_id, content=content)
except AnansiError as e:
print(f"Anansi API Error (Status {e.status_code}): {e.args[0]}")
if e.status_code == 401:
# Handle invalid API key
pass
elif e.status_code == 402:
# Handle monthly quota exceeded (don't retry)
pass
elif e.status_code == 429:
# Handle rate limit exceeded (retry with backoff)
time.sleep(60)
elif e.status_code == 413:
# Handle content too large
pass
else:
# Handle other client or server errors
pass
Rate Limiting and Quotas
The API enforces both rate limits (requests per minute) and monthly quotas.
- Rate Limits (
429): If you exceed the rate limit, the API will return a429 Too Many Requestserror. The SDK will raise anAnansiErrorwith this status code. It is safe to retry these requests after a short delay (e.g., using exponential backoff). - Quotas (
402): If your monthly usage quota is exceeded, the API will return a402 Payment Requirederror. These requests should not be retried automatically. Instead, this indicates that you need to upgrade your plan.
Idempotency
The ingest method supports a source_id parameter. You can use this to provide a unique identifier for each piece of content you ingest. If you try to ingest content with a source_id that has already been used, Anansi will recognize it as a duplicate and will not re-ingest the content, effectively making the ingest operation idempotent for that source_id. This is useful for preventing duplicate memories from being created due to retries or redundant processing.
# If this is called multiple times, the content will only be ingested once.
memory.ingest(
user_id="user_123",
content="This is a unique piece of information.",
source_id="my-unique-source-id-123",
)
Pagination
The list_memories method supports pagination through limit and offset parameters. It returns a tuple of (memories, total).
page_size = 20
current_page = 0
total_memories = 0
while True:
memories, total = memory.list_memories(
user_id="user_123",
limit=page_size,
offset=current_page * page_size,
)
if not memories:
break
total_memories = total
print(f"Page {current_page + 1}: Found {len(memories)} of {total} total memories.")
current_page += 1
print(f"Finished paginating through all {total_memories} memories.")
Requirements
Python 3.9+. No external dependencies — uses stdlib urllib only.
Links
MIT licensed.
Metadata
Release files for anansi-memory 0.3.1
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 |
|---|---|---|---|---|
| anansi_memory-0.3.1.tar.gz | 7.9 kB | With provenance | Details for anansi_memory-0.3.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| anansi_memory-0.3.1-py3-none-any.whl | Python 3 | none | any | Details for anansi_memory-0.3.1-py3-none-any.whl |
Total release size: 16.2 kB
Release files / anansi_memory-0.3.1.tar.gz
| Download URL | anansi_memory-0.3.1.tar.gz |
|---|---|
| Size | 7.9 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
82633579760998f007ea6a25e37ee093aa4c4095bffd8b356b91791abb3c87ab
|
|
BLAKE2b-256 checksum How to use checksums |
8323c4c7c84ef0ac63ee4789701a99027c73cea6480647e1bc6c6ad5aacf052d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
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.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 5, 2026.
Transparency logRelease files / anansi_memory-0.3.1-py3-none-any.whl
| Download URL | anansi_memory-0.3.1-py3-none-any.whl |
|---|---|
| Size | 8.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
e2e7810573d9f5ddf7e918886b3edac2673fd9553d8256747a2be0b7b1af70f9
|
|
BLAKE2b-256 checksum How to use checksums |
4c8a4b4f2c63bb52e24b3fc790cf2b09e1d77435162671bb230872cd7802c9f3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
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.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 5, 2026.
Transparency logSecurity for anansi-memory 0.3.1
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.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for anansi_memory-0.3.1.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Aug 5, 2026.
Transparency logA 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.