Matt Aitchison be566f5e05 fix(tracing): defer the auth token read from import to first backend call ...

The module-level EventListener builds a TraceBatchManager, whose __init__ called get_auth_token(). That instantiated TokenManager, which runs Fernet.generate_key() and writes secret.key to the user data dir, so `import crewai` consumed OS entropy and touched disk even with tracing disabled. Runtimes that forbid randomness during import snapshots fail on this.

plus_api is now a cached_property that builds the client on first access. A missing or unreadable token (corrupt tokens file, unwritable data dir) yields an anonymous PlusAPI, since that failure now happens mid-run. The `not self.plus_api` guards were always false and are gone, along with a duplicate ephemeral_trace_url assignment.

2026-10-13 15:13:10 -06:00

..

cli

feat(cli): crewai eval --models says at once where a stopped model broke (#8105)

2026-10-13 14:58:03 -03:00

crewai

fix(tracing): defer the auth token read from import to first backend call

2026-10-13 15:13:10 -06:00

crewai-core

fix(core): give trigger list and payload requests a 60s timeout (#8081)

2026-10-12 18:55:48 +00:00

crewai-files

feat: bump versions to 1.15.27 (#8001)

2026-10-09 19:25:10 -03:00

crewai-tools

fix(utilities): stop json-schema models warning about BaseModel-shadowing fields (#8085)

2026-10-13 17:29:09 +00:00

devtools

Refresh the LiteLLM catalog after the version-bump PR merges. (#8103)

2026-10-12 23:57:15 +05:30