Skip to main content

Runtime 0.7.0: one inspectable, repairable Qdrant policy

Runtime 0.7.0 makes a new installation and an in-place upgrade converge on the same Qdrant policy. The default stays dense retrieval with float32 vectors. An upgrade does not move a collection, re-embed memories, change memory identity or change recall and write semantics.

The release adds a Runtime policy record for each profile destination. The record states the target definition, the accepted effective policy and the observed payload indexes. It is separate from point payloads. The existing schema_version=1 and the embedding identity sentinel are not migrated or rewritten.

Operator upgrade commands

Both operator entries share one implementation:

qdrant-memory --home PROFILE upgrade plan
qdrant-memory --home PROFILE upgrade apply --plan PLAN_ID
qdrant-memory --home PROFILE upgrade resume --plan PLAN_ID
qdrant-memory --home PROFILE upgrade status

hermes qdrant-memory upgrade plan
hermes qdrant-memory upgrade apply --plan PLAN_ID

plan and status never initialize a collection, write the Ledger, run recovery, create an index or change configuration. plan saves one control plane record so apply can reference it. Add --dry-run to inspect without saving.

apply and resume take the existing writer lease and re-validate the saved plan against the live profile. A stale plan is refused. A destination with pending reset, scoped deletion, shared receipt or failed event work is refused until the existing recovery path clears it.

Payload index policy

The managed index set comes from one definition that init, doctor, upgrade plan and upgrade apply all read. Existing collections keep their indexes. Runtime no longer backfills them during normal startup, a plain init or an MCP connection.

upgrade apply creates only missing managed indexes. A conflicting index type is a blocker with an operator procedure. An extra index outside the policy is reported and never removed. Embedded storage reports payload indexes as not applicable instead of simulating them.

Recovery and rollback

An embedded upgrade copies the closed local store, the Ledger, the delete and reset receipts and the profile configuration into a recovery point. It then reopens the copy and compares the point count before it continues.

A remote upgrade requires --confirm-writers-isolated and --accept-operator-backup. Without both flags the Runtime refuses the apply. This release does not hold Qdrant administrative backup rights.

An application version rollback is not a storage downgrade. Do not open data written by a newer server with an older server or an older embedded client. Restore the whole recovery point taken at the same stopped-writer boundary. Writes that landed after that point are lost by design.

Qdrant versions

Qdrant Server 1.20.1 is the recommended and rehearsed self-hosted baseline. The controlled Runtime environment pins qdrant-client==1.20.0. The published dependency range stays qdrant-client>=1.15,<2, so a bare PyPI install can resolve another client within that range. Only the controlled environment and the release acceptance run are exact pins.

The upgrade rehearsal keeps the 1.20.0 image as a fixture. It upgrades one volume in place, copies the stopped volume and restores it into a fresh volume. It then compares points, payloads, vectors and indexes.

Server and Cloud are inspected for version, capability and permission only. An unknown version reports unknown and its limited capability. Embedded keeps its local backend and has no independent server.

Not in this release

Cross-collection copy or re-embed, alias switching, BM25 or RRF recall, MMR, recency ranking, ACORN, tenant or HNSW graph control, quantization and sparse backfill are future work. Diagnostics may report a probed capability. Nothing is enabled automatically.

Installation

Check PyPI for availability before installing in a private environment:

python -m pip install 'hermes-plugin-qdrant-memory[mcp]==0.7.0'

For Hermes Catalog installations, wait for the separately reviewed Catalog repin, run hermes plugins update qdrant-memory, and restart Hermes. A Runtime publication does not update the Catalog automatically.

Upgrade notes

From 0.6.3 or 0.6.2, keep the same profile, collection and embedding pipeline. An existing destination keeps serving its dense path and is reported as legacy until an operator applies a plan. Run upgrade plan, review the blockers and the index status, then apply inside a short maintenance window.

Catalog users upgrading from 0.5.4 must keep the same embedding pipeline to reuse a collection. The optional CPU FastEmbed extra changed from 0.9.0 to 0.9.1, which changes its embedding fingerprint even with the same model and dimensions. Preserve the old environment and data. Do not infer reuse from matching dimensions and do not overwrite the identity sentinel.

See the 0.6.3 release for the preceding Runtime changes.

Metadata

Release files for hermes-plugin-qdrant-memory 0.7.0

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

Source distribution (sdist)

Source distribution for hermes-plugin-qdrant-memory 0.7.0
File Size Uploaded Provenance Actions
hermes_plugin_qdrant_memory-0.7.0.tar.gz 475.7 kB With provenance Details for hermes_plugin_qdrant_memory-0.7.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for hermes-plugin-qdrant-memory 0.7.0
File Interpreter ABI Platform Actions
hermes_plugin_qdrant_memory-0.7.0-py3-none-any.whl Python 3 none any Details for hermes_plugin_qdrant_memory-0.7.0-py3-none-any.whl

Total release size: 656.1 kB

Release files / hermes_plugin_qdrant_memory-0.7.0.tar.gz

Download URL hermes_plugin_qdrant_memory-0.7.0.tar.gz
Size 475.7 kB
Tags Source
SHA-256 checksum
How to use checksums
4bd4653e02b2b40c141e7bf46947205f4409844cad8958cb6abda18d8f7a834b
BLAKE2b-256 checksum
How to use checksums
5035eaed53c7afcd2db690d1e1c233c62bb7d86fdd5eb6f85135eed820fbe86e
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

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 Nov 14, 2026.

Transparency log

Release files / hermes_plugin_qdrant_memory-0.7.0-py3-none-any.whl

Download URL hermes_plugin_qdrant_memory-0.7.0-py3-none-any.whl
Size 180.4 kB
Tags Python 3
SHA-256 checksum
How to use checksums
931737e5ce30e5e708144a3fe88622f1daedbe233006c8b6a736623f72511ffa
BLAKE2b-256 checksum
How to use checksums
7e8b1fc14147ccbadc537bc18606eb08c8067b6a320e27eaa5c05b470ee516cf
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

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 Nov 14, 2026.

Transparency log

Security for hermes-plugin-qdrant-memory 0.7.0

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.

Select from 2 files

Attestation for hermes_plugin_qdrant_memory-0.7.0.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 Nov 14, 2026.

Transparency log

A 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.

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