crewAI 1.15.42
- Source
- github.com
- Author
- Vidit-Ostwal
- Date

Context
crewAI 1.15.42 changes where an 's Knowledge lives, and existing users have to re-index. Knowledge is the crewAI feature that loads documents, such as PDFs or text files, into a vector database so agents can search them by meaning while they work. Until this release, that database was ChromaDB by default. Now it is Qdrant, running locally from a folder on disk with no server, and ChromaDB becomes an optional extra installed with `pip install "crewai[chromadb]"`.
The switch does not carry old data over. The Knowledge documentation for 1.15.42 says Knowledge indexed with ChromaDB in an earlier version "is not read automatically, and nothing is migrated." A project that upgrades either runs its crew again, which re-embeds every source and so calls the model again, or opts back in to ChromaDB by installing the extra and calling `set__config(ChromaDBConfig())`.
The new default also does not keep Knowledge between runs. Each process gets its own temporary Qdrant folder, named after its process ID, and deletes it on exit, so separate workers never collide but every run embeds its sources again. To keep an index on disk, a project sets a fixed path with `QdrantConfig(options={"path": "./knowledge_db"})`, and only one process can use that folder at a time. Teams that need one shared store across processes or replicas can point crewAI at a Qdrant server or go back to ChromaDB. The rest of the release adds one smaller agent feature and a list of bug fixes, including a raised minimum version of the soupsieve library to address a listed security vulnerability.
- crewAI now honors max_execution_time without waiting for the worker to finish.
- Standalone agents can now have tools picked for them and more tools added later.
- output_json values are now stored as JSON, fixing date issues in .json output and output_file.
- The soupsieve dependency floor rises to 4.0 to address CVE-2026-126190.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
- RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.
Checking sign-in…
Loading comments…


