Skip to content

Talos Memory 0.2.0a1

Pre-release
Pre-release

Choose a tag to compare

@Musyg Musyg released this 04 Oct 18:51

Talos Memory is an alpha toolkit for versioned agent memory and explicit policy evidence.

This release contains two independently installable distributions:

  • hermes-memory-core 0.2.0a1: immutable source versions, exact dependency references, revocation, restore quarantine, CLI, local MCP, optional authenticated HTTP and experimental typed lexical retrieval.
  • hermes-policy-registry 0.1.0: immutable policies and evidence, explicit proposal/evaluation/admission/rollback states, and a synthetic withdrawal demonstration. Admission does not execute a policy or grant runtime permissions.

The repository includes English and French documentation, synthetic examples, 158 distinct tests and an installed-distribution CI recipe. Local qualification covered Linux Python 3.12, both wheels, wheels rebuilt from their source distributions, and installation together. Test repetitions are not counted as additional unique tests.

The source selection has a fresh public Git history and documented file provenance. It contains no private memory corpus, embeddings, production database, credentials or operational traces.

Install the memory core from this release:

python -m pip install https://github.com/Musyg/talos-memory/releases/download/v0.2.0a1/hermes_memory_core-0.2.0a1-py3-none-any.whl
hermes-memory-demo --directory ./new-memory-demo

The optional registry has its own wheel and source archive. SHA256.json lists artifact checksums.

Typed retrieval is experimental: it improved evidence selection in a bounded synthetic comparison but failed its registered relative latency gate. This release does not claim a general speed improvement, model unlearning, physical erasure or protection when authority and freshness witnesses are rolled back together. Read the recovery and service trust-boundary documentation before integration. No model, external database or provider account is required for the examples.