generated: '2026-09-05'
method: searched
source: https://github.com/cognizant-ai-lab/neuro-san/releases
docs: https://github.com/cognizant-ai-lab/neuro-san/releases
note: >-
  Cognizant AI Lab publishes dated, per-release change notes as GitHub Releases on the
  neuro-san repository. There is no CHANGELOG.md in the repository (probed, 404) and no
  changelog page on cognizant.com — the Releases feed is the changelog.
  Entries below are the ten most recent, read from the repository's releases Atom feed on
  2026-09-05. Highlights are drawn verbatim from the provider's own notes; where the notes
  are a bare list of merged pull requests, that is what is recorded, and no editorial
  summary is invented.
scheme: semver-shaped (0.MAJOR.MINOR)
current_version: 0.7.0
current_version_date: '2026-09-03'
release_count_total: 149
format: github-releases
machine_readable_feed: https://github.com/cognizant-ai-lab/neuro-san/releases.atom
breaking_change_labeling:
  present: false
  detail: >-
    Notes are auto-generated "What's Changed" PR lists. Only 0.7.0 carries a hand-written
    "Important" callout. There is no breaking/additive labeling a machine can filter on.
entries:
- version: 0.7.0
  date: '2026-09-03'
  breaking: unclear
  highlights:
  - >-
    Important callout — "Requires Python 3.12 or above. That was already indirectly the case
    since 0.6.84." Formalizes an existing floor rather than introducing a new one.
  - Make neuro-san formally require Python 3.12 or above (#1302)
  - Run unit tests on Python 3.12, 3.13 and 3.14 (#1304)
  - Add Claude Fable 5.1 to the supported LLM set (#1297)
  - Accept an LLM endpoint base URL without /v1 in the record/playback proxy (#1296)
  - Capture root cause of HTTP load-test failures (#1300)
  - Add a flag to allow prompt caching in load test (#1295)
- version: 0.6.98
  date: '2026-08-26'
  breaking: false
  highlights:
  - Use the shared full-text projection at both capture points (#1285)
  - Run EventInitiator in only one of the neuro-san service instances (#1289)
- version: 0.6.97
  date: '2026-08-24'
  breaking: false
  highlights:
  - Unify responses handling and refactor tests (#1278)
  - Add a ContentUtils policy class and block-content test infrastructure (#1276)
  - Flatten list content to full text in the wire converter (#1281)
  - Fix duplicate descendant journaling events (#1259)
  - Re-enable the Ollama nightly test case (#1284)
- version: 0.6.96
  date: '2026-08-21'
  breaking: false
  highlights:
  - Add a record/playback LLM proxy for load and regression testing (#1261)
  - Add instance id to resources snapshot data (#1274)
  - Several transpilation-related changes (#1270, #1273, #1275)
- version: 0.6.95
  date: '2026-08-19'
  breaking: unclear
- version: 0.6.94
  date: '2026-08-14'
  breaking: unclear
- version: 0.6.93
  date: '2026-08-14'
  breaking: unclear
- version: 0.6.92
  date: '2026-08-11'
  breaking: unclear
- version: 0.6.91
  date: '2026-08-08'
  breaking: unclear
- version: 0.6.90
  date: '2026-08-06'
  breaking: unclear
entries_note: >-
  Versions 0.6.90 through 0.6.95 carry release notes in the feed (2,000-8,400 characters each)
  but were not transcribed here; the window above is deliberately recent. Date and version are
  recorded for all ten so the cadence is measurable, and `breaking: unclear` states honestly
  that the notes were not read closely enough to rule a breaking change in or out.
related_feeds:
- name: nsflow releases
  url: https://github.com/cognizant-ai-lab/nsflow/releases
- name: neuro-san-studio releases
  url: https://github.com/cognizant-ai-lab/neuro-san-studio/releases