Metadata-Version: 2.4
Name: hermes-plugin-qdrant-memory
Version: 0.6.0
Summary: Durable Qdrant memory for Hermes, standalone CLI and local MCP
License-Expression: MIT
Project-URL: Homepage, https://github.com/cnkang/hermes-plugin-qdrant-memory
Project-URL: Repository, https://github.com/cnkang/hermes-plugin-qdrant-memory
Project-URL: Issues, https://github.com/cnkang/hermes-plugin-qdrant-memory/issues
Project-URL: Changelog, https://github.com/cnkang/hermes-plugin-qdrant-memory/blob/main/CHANGELOG.md
Project-URL: MCP, https://github.com/cnkang/qdrant-memory-mcp
Project-URL: Skill, https://github.com/cnkang/skills/tree/main/qdrant-memory
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: qdrant-client<2,>=1.15
Requires-Dist: httpx<1,>=0.28
Requires-Dist: portalocker<4,>=3.2
Requires-Dist: jsonschema<5,>=4.23
Requires-Dist: pywin32<313,>=311; sys_platform == "win32"
Provides-Extra: mcp
Requires-Dist: mcp<3,>=1.30; extra == "mcp"
Provides-Extra: remote
Requires-Dist: mcp<3,>=1.30; extra == "remote"
Requires-Dist: uvicorn<1,>=0.35; extra == "remote"
Provides-Extra: local
Requires-Dist: fastembed==0.9.1; extra == "local"
Dynamic: license-file

# Runtime 0.6.0: portable Codex memories

This release adds explicit Codex v1/v2 semantic-memory inspection and import,
plus readable, verifiable JSON/Markdown handoff packages. Native sources and
Qdrant records remain independent of plugin installation. Uninstalling the
plugin does not remove stored memories or exports.

Inspect the installed `qdrant-memory codex --help` before selecting an operation.
Preview imports and review skipped records before committing them. Credentials,
code fences, raw conversations and unsupported artifacts do not become semantic
memories through this adapter.

Handoff verification checks record identity, content, scope, size and checksums.
It does not claim that Codex has adopted the package: the status remains
`HANDOFF_PENDING` until the destination confirms actual adoption. Bulk native
writeback is not provided.

macOS/Linux use descriptor-relative no-follow reads. Windows uses checked native
handles, refuses reparse points and remote volumes, and requires persistent
private ACLs for created handoff directories. Unsupported filesystem guarantees
fail closed.

Scoped deletion also supersedes quarantined events for the deleted scope.
Retrying failed work or restarting the provider cannot recreate those deleted
memories; records in other scopes remain intact.

The optional CPU backend uses upstream FastEmbed 0.9.1. Profiles created before
Runtime 0.5.5 may record the older 0.9.0 pipeline fingerprint. Preserve their
environment, store and cache; export authorized records and re-embed them into
a new profile and collection before switching. Do not relabel an existing
collection's pipeline fingerprint. See the CPU setup guide below.

Explicit model preparation allows 15 minutes per artifact, retaining byte
limits, immutable hashes and atomic cache publication. Slow networks have a
wider window. An incomplete or over-budget download never becomes a ready cache.
All model downloads share a 30-minute network budget. Real CPU CI allows
60 minutes, leaving time for dependencies and acceptance after download.
The cancellable deadline covers connection, response headers and streamed body
reads, including continuous tiny fragments. Timeout closes the stream/client,
removes temporary files and releases the preparation lock before a retry.

The Codex/Ollama Cloud acceptance uses synthetic data and local embedding
fixtures. It validates harness/tool behavior, not embedding-model inference.
Runtime, Python MCP, managed launchers and Skills have independent releases and
reviewed pins. Publishing this Runtime does not automatically upgrade them or
publish the Codex Marketplace candidate.

## Install this release

Confirm version 0.6.0 is available on PyPI before installing:

```sh
python -m pip install 'hermes-plugin-qdrant-memory[mcp]==0.6.0'
qdrant-memory --help
```

This installs the standalone CLI and MCP base without Hermes. Add the `local`
extra for optional CPU embeddings. Model preparation requires explicit approval.
Configure a private profile before memory operations. See the
[standalone guide](https://github.com/cnkang/hermes-plugin-qdrant-memory/blob/v0.6.0/docs/standalone.md)
and [CPU setup](https://github.com/cnkang/hermes-plugin-qdrant-memory/blob/v0.6.0/docs/local-embedding.md).
