loading
loading
Replit Acquires ChromaDB and Shuts Down Chroma Cloud in 60 Days Replit is acquiring the vector database maker to build what both companies call the context layer for autonomous enterprise agents. - Replit is acquiring Chroma, the open source vector database used widely for RAG, to build an agent context layer. - Deal terms were not disclosed; Replit was last valued at $9B after a $400M Series D led by Georgian in March 2026. - Chroma Cloud is winding down over 60 days; customers must migrate off the hosted service. - Open source ChromaDB continues with 6M monthly downloads and additional private code open sourced under Apache 2.0. - Replit CEO Amjad Masad and President Michele Catasta were early Chroma investors, dating the relationship to 2023. - Signals broader consolidation of vector databases into AI coding and agent platforms like Replit, Cursor and Cognition. Replit is acquiring Chroma, the company behind the widely used open-source vector database ChromaDB, according to the companies’ acquisition announcement. Financial terms were not disclosed. Replit plans to combine Chroma’s retrieval and memory infrastructure with its platform for building and running software agents. Chroma Cloud will wind down over 60 days, requiring customers of the managed service to migrate their production workloads. Self-hosted ChromaDB remains available under the Apache 2.0 license, and Replit has committed to contributing upstream. Chroma CEO Jeff Huber says ChromaDB and the company’s newer Foundation product will continue inside Replit. Replit buys its memory layer Replit Agent turns natural-language instructions into working software, handling tasks such as writing code, configuring services and deploying applications. Those agents need access to relevant documents, source code, tickets, messages and internal tools before they can complete broader company workflows. Chroma supplies infrastructure for finding that information. A vector database stores embeddings, which are numerical representations of text, images or other data. Retrieval-augmented generation, commonly called RAG, searches those representations and places relevant results into a model’s prompt before the model responds or acts. Huber describes the combined product as a “context layer for the self-driving company.” In practical terms, Replit wants its agents to retrieve company knowledge, retain useful information across tasks and apply that context while executing workflows with less manual guidance. Replit says its platform has more than 60 million users and is used by people at 85% of Fortune 500 companies. It names Atlassian, Adobe, Visa and Databricks among its customers. In March 2026, the company raised $400 million in a Series D led by Georgian at a $9 billion valuation, three times its valuation six months earlier. Replit’s relationship with Chroma predates the transaction. Replit President Michele Catasta and CEO Amjad Masad invested in Chroma in 2023, and the teams have collaborated since that year. That history gave Replit direct experience with Chroma’s technology before pursuing the acquisition. Chroma brings more than vector search Chroma released ChromaDB in early 2023 following ChatGPT's release, shifting the company's focus from training to inference and solving the problem of context for agents. Huber says Chroma once estimated that it accounted for more than 95% of vector-search downloads. He names SpaceX, xAI, Mintlify, Krea, Paramount, Retool and CoreWeave among organizations using its technology. At the time of the acquisition announcement, Chroma reported nearly 30,000 GitHub stars and six million downloads per month. Its engineering work had expanded beyond the original local vector store into infrastructure for larger, distributed deployments: - Distributed and serverless architecture: systems designed to spread workloads across machines and provision capacity on demand. - wal3: a write-ahead log built on Amazon S3 that records changes for durable storage and recovery. - Object-storage-native retrieval: search infrastructure designed to work with data held in object storage. - Custom query execution: an engine for planning and running vector searches, filters and related retrieval operations. Chroma launched Foundation in mid-2026 as a system for building organizational context that can improve through use. Huber says it achieved state-of-the-art results on the BEAM memory benchmark at launch. The announcement provides no independent benchmark analysis, deployment requirements or pricing details for Foundation. The cloud deadline is 60 days Huber said Chroma will “responsibly” wind down its existing cloud service over the 60 days following the announcement. Customers running production workloads on Chroma Cloud need to confirm their exact shutdown date and move their data and applications before access ends. A production migration should preserve collection identifiers, embedding dimensions, embedding models, metadata, filters and application-level access controls. Teams should test retrieval quality and latency against representative queries, update endpoints and credentials, and verify backups before directing all reads and writes to the replacement system. Apache code keeps moving Self-hosted ChromaDB deployments remain under their operators’ control and have no dependency on the managed Chroma Cloud service. Chroma says it has moved additional code from private repositories into the Apache 2.0 project to support continuity. The Apache 2.0 license allows developers to use, modify and fork the published code. Replit has also committed to contributing upstream as it extends Chroma for enterprise context and memory workloads. The announcement omits several governance details that matter to maintainers and production users, including future release cadence, maintainer authority, long-term compatibility guarantees and the boundary between public ChromaDB features and proprietary Replit services. Replit pulls more of the agent stack inside Replit previously acquired OpenInt to strengthen connectors, which let agents reach external services and company data. Chroma adds retrieval, storage and memory to that stack, giving Replit control over more of the path from data access to agent execution. Owning those components can help Replit tune retrieval and agent behavior together, reduce integration work and package security controls for enterprise deployments. It also creates platform-coupling considerations for teams that want to move workloads among agent runtimes, model providers and vector databases. The acquisition removes a prominent developer-focused vector database company from the independent market. Pinecone, Weaviate, Qdrant, Turbopuffer and LanceDB remain among the vendors competing for retrieval workloads, while Chroma’s commercial development will increasingly reflect Replit’s agent platform and enterprise requirements. The integration details still missing - What is the exact final service date for each Chroma Cloud customer? - Which migration tools, export formats and support options will Chroma provide? - How will Foundation be packaged and priced inside Replit? - When will Replit expose Chroma-backed memory and retrieval APIs? - Which data residency, retention, isolation and audit controls will enterprise deployments support? - How will Replit govern ChromaDB releases and accept outside contributions?
document fetched 2026-10-24 15:22