LQ.AI v0.7.1 is a patch release centred on two improvements: a refreshed OpenWebUI foundation and chat attachments that now make it all the way to the model. It also brings dependency updates, clearer contributor and deployment documentation, and a more reliable release process.
There are no breaking changes or required configuration changes in this release.
Highlights
OpenWebUI moves to v0.11.0
The vendored OpenWebUI fork has been refreshed from v0.9.2 to v0.11.0. This is the first quarterly refresh completed with the new rebase runbook, which gives us a repeatable process for keeping the fork current. Web dependency automation is now limited to security updates, reducing avoidable conflicts with future upstream refreshes. (#541, #539, #542)
Chat attachments now reach the model
Files attached in the chat composer are now included with the message sent to the model. The interface shows whether each file is still processing, ready, or failed; failed files are left out of the message; and uploads are cleaned up safely when you switch chats. The client also checks the 16-file limit before sending. (#402)
Safer release ranges
Release automation now selects the previous LQ.AI tag only from tags reachable from the commit being released. This prevents vendor tags such as OpenWebUI's version tags from accidentally widening a release range. (#552)
More improvements
- Gateway tests now share a single example-environment fixture and are isolated from values in a developer's shell. (#497, #527)
- Contributor documentation, Make targets, and agent onboarding now match the checks that CI actually runs. (#505)
- A new deployment guide explains how to connect the gateway to an Ollama instance over a tailnet-hosted HTTPS endpoint. (#510)
- ADR 0026 is now accepted: Docling is out of near-term scope, while a pluggable parser interface remains deferred. (#528)
- Project documentation now records the completed AG-01 retrieval-ownership work. (#496)
Dependency and workflow maintenance
- Routine grouped updates landed across the API and gateway, including Pydantic Settings, Starlette, Alembic, Greenlet, PyMuPDF, Tiktoken, spaCy, Ruff, and mypy. (#550, #543, #513, #500)
- The gateway now uses
cryptography50.0.0, and its Starlette range has been aligned with the API. (#516, #514) - Uvicorn was updated in both the API and gateway, SQLAlchemy in the API, and
pymdown-extensionsin the web fork. (#522, #517, #484, #538) - LangGraph is no longer included in routine API dependency groups, so its upgrades can be handled on their own cadence. (#546)
- The release workflow now uses the current
docker/login-action. (#478)
Upgrade notes
The LQ.AI API schema is unchanged at Alembic revision 0066_backfill_project_knowledge_bases.
The OpenWebUI refresh includes 15 upstream Alembic migrations. They run automatically when the web service starts, so there is no manual migration step. If you have a long-lived OpenWebUI database, take a backup before upgrading and allow a little extra time for the first start.
Current limits
- Chat attachments are kept in the current page's draft state and are not restored after a page reload.
- The desktop launcher still defaults to the
latestimage tag. SetLQ_AI_IMAGE_TAGin.envwhen you need a reproducible, pinned stack. - The macOS desktop build is currently available for Apple silicon only.
Contributors
Thank you to @Bhoomi-jain, @katsugtgz, @pscripps, and @SaifAlYounan for their contributions to this release, alongside maintainer work and Dependabot updates.
Downloads
- macOS Apple silicon:
LQ.AI-0.7.1-arm64.dmg - Container images:
ghcr.io/legalquants/lq-ai-api:0.7.1,ghcr.io/legalquants/lq-ai-gateway:0.7.1,ghcr.io/legalquants/lq-ai-web:0.7.1, andghcr.io/legalquants/lq-ai-proxy:0.7.1
See the full changelog for all 25 merged pull requests.
Range: v0.7.0..v0.7.1 (b092e1f5b6b14cfbc1440b1f563e37941731ed7f..bc8b639c8af92ca11b79fb222f9bd9a0b512e969). Drafted with lq-maintainer-agent v0.6.0 and reviewed by @houfu.
