audit-ready-memory
Audit-ready, local-first memory for AI agents: a time-stamped log you can read, replay, and delete from, instead of hidden state inside a model or a hosted service.
Memory is stored as an append-only sequence of events in an embedded ArcadeDB database. Nothing is ever updated in place. Reads are recorded as well as writes, so you can answer "what did the model actually see, and when?" Deletion is an appended tombstone with a required reason, and deleted content can never re-enter a model's context.
Funded by SIDN Fund.
Install
The package on PyPI is not current yet; install from source until v0.1.0 is released:
git clone https://github.com/humemai/audit-ready-memory
cd audit-ready-memory
uv sync
Python 3.10 or newer. The database is embedded, so there is no server to run.
Thirty-second example
from audit_ready_memory import Memory, Message, Document
with Memory("./memory-db") as memory:
memory.add_message(Message(speaker="anna", content="the water pump failed on tuesday"))
memory.add_document(Document(name="repairs.txt", text="pump replaced on wednesday"))
# Search, and record the search in the same step.
recall, hits = memory.recall("pump")
for hit in hits:
print(hit.seq, hit.kind)
# The read is now in the log too.
print(len(memory), "events")
Every event carries an id, a UTC timestamp, and a seq assigned by the store. seq
gives the log a total order that does not depend on the clock.
What you get
| Capability | How |
|---|---|
| Read the whole history | memory.events(), filterable by kind, session, or seq range |
| Rebuild any past state | memory.replay(up_to_seq=n), a pure fold over the log |
| See what the model saw | every read appends a Recall event naming the exact items |
| Delete with a reason | memory.delete(ids, reason="...") appends a tombstone |
| Run it anywhere | embedded database, no server, no network |
Four kinds of event
Messageis one conversational turn. Who said what.Documentis data someone uploaded, stored as extracted text.Recallis a read. It names which stored items entered the model's context, and for which query.Deletionis a tombstone naming its targets and why they were removed.
Read the memory model for the invariants these guarantee.
Demo
A Streamlit app showing a small group chatting with each other and an AI, sharing documents, on top of an audit-ready memory:
uv run --extra demo streamlit run demo/app.py
Without an OPENROUTER_API_KEY the app still runs. Browsing, the audit log, replay, and
deletion all work. Only the AI chat is disabled.
Honest limits
- Deletion is a tombstone, not erasure. Deleted content stays in the append-only log and is excluded from every read. If you need the bytes gone from disk, this is not that.
retain_untilis a marker, not a policy. The field is recorded but nothing enforces it yet; an expired event is still visible. Track it or set it, but do not rely on it.- Retrieval is keyword overlap. Deliberately simple and deterministic. No embeddings.
- The demo's model call is a cloud call. The memory is local; recalled text sent to OpenRouter is not.
More in limitations.
Documentation
https://docs.humem.ai/audit-ready-memory/latest/
Development
uv sync --all-groups --all-extras
uv run pytest
License
MIT. See LICENSE.
Metadata
Release files for audit-ready-memory 0.0.2
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 |
|---|---|---|---|---|
| audit_ready_memory-0.0.2.tar.gz | 184.1 kB | With provenance | Details for audit_ready_memory-0.0.2.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| audit_ready_memory-0.0.2-py3-none-any.whl | Python 3 | none | any | Details for audit_ready_memory-0.0.2-py3-none-any.whl |
Total release size: 197.6 kB
Release files / audit_ready_memory-0.0.2.tar.gz
| Download URL | audit_ready_memory-0.0.2.tar.gz |
|---|---|
| Size | 184.1 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
d77c76141f800bc65ef5e258f6e2615830f883a1d5a2458bb0b76eaf95d9b399
|
|
BLAKE2b-256 checksum How to use checksums |
5b776d701dbf5d4f922d072528fb51f81020c920a53acfba41b8391a2419fad2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.7
|
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 14, 2026.
Transparency logRelease files / audit_ready_memory-0.0.2-py3-none-any.whl
| Download URL | audit_ready_memory-0.0.2-py3-none-any.whl |
|---|---|
| Size | 13.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
5c2ec20567c7c21b0194ac0fce9b2ae118d15d46a6fe89b45385ad4ba842ed43
|
|
BLAKE2b-256 checksum How to use checksums |
47d0d951190aace3bdcdfb9fc28bfb1cedbe5216e80b530d147d3af28758fbe5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.7
|
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 14, 2026.
Transparency logSecurity for audit-ready-memory 0.0.2
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 audit_ready_memory-0.0.2.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 14, 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.