Skip to main content

memoriesQL

Build memories with meaning.

memoriesQL is an open-source, local-first memory foundation for AI agents. It preserves authorized evidence in PostgreSQL and supports agent-authored structured observations with inspectable source support.

The memoriesql package provides the provider-neutral open core: canonical migrations, versioned contracts and composable Python runtime mechanics. It is not the memoriesQL desktop product or a configured background memory service.

Experimental / pre-alpha. Installed fictional tests qualify mechanics, not real-provider quality or production readiness. Applications must supply source qualification, authority, worker composition and qualified model bindings.

Install and explore

Use Python 3.13 or 3.14 (>=3.13,<3.15). This page describes the 0.0.11 distribution. Until its publication is independently verified, use the reviewed local wheel; after publication, install the exact pin below. Published 0.0.10 remains available without authored claims and relations (schema 27).

python3.13 -m pip install memoriesql==0.0.11
memoriesql --version
memoriesql contracts --json
memoriesql contract memoriesql.capture.connector-cursor --json
from memoriesql.contracts import get_contract, load_catalog

schemas = load_catalog("json_schema")
cursor_contract = get_contract("memoriesql.capture.connector-cursor")

Catalog inspection uses bundled resources without a database, provider credentials or external service access. The CLI does not start a memory service.

What you can build on

  • Durable authorized raw capture, receipt-backed progress and retained-fold recovery.
  • Stable qualified source identity and immutable evidence packages.
  • A durable semantic-task queue with fenced execution, cancellation and accounting.
  • Structured observations, explicit distinct-bead corrections, local entity mentions and attributable classification.
  • Authorized source revisiting and stored-memory/evidence inspection without inference on reads.

A bead is a source-anchored observation whose statements carry meaning, attribution, conditions and uncertainty. Its short title and summary are human navigation, not replacement evidence. Meaning is authored by an agent or human; deterministic code validates and persists it.

Version 0.0.11 adds authored claims and relations (schema 27, author-complete-unit revision 6): the primary author of one complete unit may optionally propose tracked claims, evidence-backed relations to explicitly pinned candidate beads and lifecycle judgments about those candidates' claims, and assess every pinned candidate. Code validates identity, pins and evidence; it never infers a relation, conflict or winner. Claim and relation states are derived from append-only history as of a known time, and a derivative never counts as independent corroboration. Dependencies are unchanged from 0.0.10. Version 0.0.10 binds authored statements to the executor's run reference before hashing and settles data-error canonical-apply refusals as invalid output. Version 0.0.9 combined declared evidence scopes (schema 25), explicit supervised managed dispatch (schema 26), and terminal-failure cleanup that retains errors without claiming a failed write persisted. Hard-bounded model admission remains the default; supervised managed dispatch requires separate exact approval and records its allowances and reported or unavailable usage truthfully. The package supplies no provider adapter or live-call permission. No live-provider semantic quality, production recall or complete-input correction/reauthoring is claimed.

Explicit composition, not automatic setup

Installation does not migrate PostgreSQL, grant access, provision trusted producers, activate tasks, start workers or configure a provider. Consumers must explicitly compose those operations and authorize any disclosure to a model. Core dependencies do not include provider extras.

Required target evidence must be supplied through the trusted execution path; mechanical exposure does not prove comprehension or whole-source completeness. Incomplete work stays incomplete. Hosts retain the worker/event loop through wait_for_cleanup(); cancellation is not proof of remote termination, rollback or reconciled usage.

Start with the repository guide, migration operations, caller composition and model admission. Development documentation may describe changes not yet published; check its release status before adopting them.

Compatibility and license

All 0.x APIs are experimental, with no general compatibility guarantee. Published contract payloads remain immutable; breaking payload changes require a new contract version, and changed published artifacts require a new distribution version. Pin releases and apply migrations explicitly.

Public core owns the canonical schema and provider-neutral runtime. Desktop surfaces, source-specific acquisition/parsing, installers, provider bindings, hosted services and sync are outside this package.

Source is Apache-2.0 licensed. The memoriesQL name and artwork remain John Inniger's trademarks under the trademark policy. Contributions use the Developer Certificate of Origin.

Release files for memoriesql 0.0.11

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for memoriesql 0.0.11
File Size Uploaded
memoriesql-0.0.11.tar.gz 563.6 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for memoriesql 0.0.11
File Interpreter ABI Platform
memoriesql-0.0.11-py3-none-any.whl Python 3 none any Details

Total release size: 1.2 MB

Release files / memoriesql-0.0.11.tar.gz

Download URL memoriesql-0.0.11.tar.gz
Size 563.6 kB
Tags Source
SHA-256 checksum
How to use checksums
80ac82ba7125efeba1d2ce172dfa97dcaabe96ab7ba451ee0fb4777a7060eae2
BLAKE2b-256 checksum
How to use checksums
a1424f1e892cef678032e5f65869c6de3ef0a642fe321d327fa4a1b47d73e61e
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 Sep 23, 2026.

Transparency log

Release files / memoriesql-0.0.11-py3-none-any.whl

Download URL memoriesql-0.0.11-py3-none-any.whl
Size 633.1 kB
Tags Python 3
SHA-256 checksum
How to use checksums
9a3ec138f54817a4b0bb3e899e5690b6ae97409581e3f233c46fa2e466b03ec7
BLAKE2b-256 checksum
How to use checksums
84295e5f67f8531425db44ee4637747b92e7462cb45dd3a7ae8ccbba7fb2a512
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 Sep 23, 2026.

Transparency log

Release history Release notifications | RSS feed

0.0.12

2 release files

This release

0.0.11 This release

2 release files

0.0.10

2 release files

0.0.9

2 release files

0.0.8

2 release files

0.0.7

2 release files

0.0.6

2 release files

0.0.5

2 release files

0.0.4

2 release files

0.0.3

2 release files

0.0.2

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page