Skip to content

Recall 0.1.0

Choose a tag to compare

@github-actions github-actions released this 11 Jun 14:13
· 96 commits to main since this release

Initial public release — an early, honest working runtime for auditable agent
memory.

Added

  • Strict recall.write.v1 admission path with firewall checks and rollback.
  • SQLite-backed graph store with addressable cells and tag-composed subgraphs.
  • Local semantic search with a deterministic hash backend.
  • Encrypted Secrets side graph with explicit save confirmation.
  • CLI, read-only TUI, and stdio MCP server.
  • Background daemon maintenance path with SQLite-backed lease control.
  • N-ary hyperedges, sandboxed hyperedge programs, DAG overlays, holonomy
    analysis, derivation closure, and eval persistence.
  • End-to-end release smoke script covering user and agent workflows.
  • Real-embedding semantic backend (mpnet:v1, 768-dim,
    sentence-transformers all-mpnet-base-v2) via the documented
    RECALL_EMBEDDING_COMMAND plug-in point. Ships two Python scripts:
    python/scripts/recall_mpnet_embedder.py (single-call adapter for
    per-query and per-write embedding) and
    python/scripts/recall_semantic_real.py (batched standalone
    indexer with query / compare / verify / status subcommands).
    Optional dependency pinned in python/requirements-semantic.txt.
    The default hash:v1 backend remains zero-dependency and unchanged.
  • Enforcement hook templates in python/hooks/: UserPromptSubmit
    injection (recall_inject_context.py.sample), Stop/SubagentStop
    writeback reminder (recall_writeback_reminder.py.sample), and
    PreToolUse guard (recall_pretooluse_guard.py.sample) for hard
    blocking of mutations without a recorded rationale cell.
  • Compliance instrumentation: python/hooks/audit_compliance.py
    (point-in-time score) and python/hooks/longitudinal_tracker.py
    (snapshot + trajectory analysis: rationale quality, rework rate,
    enforcement events, continuity value).
  • Comprehensive test suites: python/hooks/test_hooks.py (hook +
    tracker + adapter tests), python/tests/system_tests.py (10
    stress + 10 capability tests), python/tests/competitive_tests.py
    (head-to-head matrix vs 6 competitor memory architectures).
  • Enforcement strategy doc: docs/17_ENFORCING_USAGE.md covering
    the four-tier discipline (system prompt, inject hook, writeback
    hook, PreToolUse guard) with rollout guidance and longitudinal
    metrics interpretation.

Full Changelog: https://github.com/H-XX-D/recall-memory-substrate/commits/v0.1.0