pipecat-moss: release 0.2.0 from this tree with the 0.1.0 port
Merged
RA
4d ago+1097+1097+0123456789012345678901234567890123456789-589-589-012345678901234567890123456789
Part of MOS-615.
pipecat-moss has two publish paths. usemoss/pipecat-moss published 0.1.0 to PyPI on 2026-10-23 (the port to moss 1.20 and pipecat-ai 1.8) with no license metadata. This tree was at 0.0.5, never published, with the BSD-2-Clause metadata (#553) and #562. Each lacked the other's changes, and a dispatch of publish-pipecat-moss.yml here would have uploaded 0.0.5 after 0.1.0.
This PR makes this tree the one source and readies 0.2.0:
- From 0.1.0:
moss>=1.20.0,<2,pipecat-ai>=1.8,<2, theMutationOptionsandQueryOptionsre-exports, its tests, the demo fix for pipecat 1.13, and handlingLLMContextFrameonly (the user aggregator consumesLLMMessagesUpdateFrame). Its 0.1.0 changelog section comes along. - Kept from here: the license metadata and SPDX headers,
cache_path,auto_refresh, and forwarding the received frame, also after a failed retrieval, with the exception on theErrorFrame. - Passages no longer pile up: each query replaces the passages the processor added for the last one, so each processor keeps only its latest retrieval in the context instead of every turn's.
- Dependencies: the
runnerextra moves to the dev group, since only the examples use it. - Publish workflow: it refuses a version below any PyPI release (PEP 440, pre-releases included) and lets the same version rerun. Every Python 3.11 to 3.14 leg tests the built wheel, and a separate job uploads and tags only after all of them pass.
After this merges and 0.2.0 publishes, usemoss/pipecat-moss's publish.yml goes away and that repository can be archived.
🤖 Generated with Claude Code
Summary by CodeRabbit
- New Features
- Added disk caching and index loading for Moss retrieval.
- Added configurable blending of semantic and keyword scoring, defaulting to 0.8.
- Exported mutation and query options for package users.
- Bug Fixes
- Retrieval failures now report the original error, and missing credentials raise a
ValueError. - Non-context frames pass through unchanged; retrieval no longer runs on message-update frames.
- Retrieval failures now report the original error, and missing credentials raise a
- Compatibility
- Supports Python 3.11–3.14, Pipecat 1.8 or later but below 2.0, and Moss 1.20 or later but below 2.0.
- Documentation
- Updated the retrieval demo to omit RTVI integration.