Replit to Acquire Chroma, the Open-Source Database Behind Context for AI Agents

The announcement is short and direct. In his post, Jeff Huber writes that Chroma “has signed a definitive agreement to join Replit”. A definitive agreement is the binding contract between buyer and seller, the stage that follows early discussions and due diligence. The post covers the purpose of the combination, the products that carry on, the changes for Chroma Cloud customers and the future of the open-source project, which together give an unusually complete picture of the deal on its first day.

The idea that anchors the deal is the self-driving company. Chroma defines it as a company where agents do more and more of the work because they know everything the organisation knows. In that framing, the hard problem for AI agents inside a business is no longer writing code or drafting text; it is knowing the right facts at the right moment. Chroma calls the software that solves this problem the context layer, and the agreement makes building that layer the shared mission of both teams.

ChromaDB and Foundation both continue at Replit

The announcement is explicit that Chroma’s two core products carry on. Huber writes: “We’re excited to continue our work on ChromaDB and Foundation at Replit building the foundational technology for the self-driving company”. He adds that the team has a lot up its sleeves. ChromaDB is the open-source database that made Chroma’s name. Foundation is the enterprise product the company started building in the middle of 2026. Both now sit inside Replit’s broader agent platform.

Huber also explains the logic of the combination in his own words: “Chroma’s state-of-the-art infrastructure for context and memory combined with Replit’s broader agent platform technology and scaled-GTM motion are a powerful combination”. GTM is shorthand for go-to-market, the sales and distribution engine that brings a product to customers. Chroma contributes the infrastructure, the research and the team that built them; Replit contributes the agent platform and the reach to put that infrastructure in front of businesses at scale.

Few acquisitions come with a paper trail as long as this one. According to the announcement, “Michele Catasta and Amjad Masad (President and CEO of Replit respectively) were early investors in Chroma and it became increasingly obvious to us both that we could do far more together than apart”. The deal is the result of a relationship that ran for more than three years before it became a transaction.

Chroma’s own record shows where that relationship began. On 7 April 2023 the company announced an $18M seed round led by Astasia Myers of Quiet Capital. The list of angel investors in that round includes Amjad Masad of Replit, alongside founders and leaders from across the developer-tools world. The same post also announced Chroma’s pre-seed round from May 2022, led by Anthony Goldbloom of AIX Ventures, James Cham of Bloomberg Beta, and Nat Friedman and Daniel Gross of AI Grant.

A Chroma template on Replit in March 2023

The first public collaboration came ten days before the seed announcement. On 28 March 2023, Chroma announced the Chroma template for Replit, which let developers build AI applications with state and memory directly inside a Repl, Replit’s browser-based project workspace. Chroma framed the launch as “the next step towards bringing the power of AI application development to the next billion software creators”. The template handled embedding documents and queries and stored documents alongside their embeddings, so a developer could add their own data and start asking questions of it.

That template was a small project, but it shows how closely the two companies’ ideas lined up from the start. Replit’s mission, as set out in its March 2026 funding announcement, is that anyone should be able to build an app without ever learning to code. Chroma’s mission, as Huber describes it, has always been bringing the right data to AI. In 2023 those ideas met in a template. In 2026 they meet in a single company.

Chroma itself started in 2022. Huber writes that the company began by working on active learning for curating training data for deep learning, then shifted its focus from training to inference when ChatGPT was released, concentrating on “the problem of context for agents”. In January 2023 he described the enterprise opportunity to an investor in plain terms: what if you could feed a really smart person all of your company’s data and then ask them questions about it. ChromaDB was released the following month.

Replit is a buyer with momentum. Founded in 2016 around the belief that anyone should be able to build an app without ever learning to code, the company announced on 11 March 2026 that it had raised $400 million at a $9 billion valuation, which it described as a 3x increase in just six months. In the same post, Replit said it has over 50 million users, counts users from 85% of the Fortune 500, and is on track to hit $1 billion in run-rate revenue by the end of 2026.

Those users range widely. Replit’s funding post describes kids building apps for school projects and their own video games, employees building internal tools from legal assistants to sales leaderboards, and enterprises using the platform for rapid prototyping and product development. India is Replit’s second-largest market, where the company recently launched partnerships with Razorpay for in-app payments and Hexaware for secure enterprise deployment, adding to its work with Humain in the Middle East.

Agent 4 puts parallel agents at the centre

Replit launched Agent 4 on the same day as its funding round, calling it its fastest and most versatile agent yet. The company says Agent 4 helps users ship production-ready software 10x faster by taking on the tedious-but-necessary work in the background. Its headline capability is parallel work: multiple agents working at the same time, each focused on a different part of the project, alongside a design canvas for exploring and refining interface variants inside a working app.

Parallel agents raise the importance of shared context. When several agents work on one project, each needs the same understanding of the code, the design decisions already made and the way the team works. That is the problem Chroma has spent four years solving, and it is why Chroma describes the combination as a powerful one. Replit has kept building out its platform in recent weeks too, announcing a Databricks integration with native Lakebase support on 10 September 2026 and a Replit Solutions Partners programme on 13 October 2026.

Replit also has a broad partner ecosystem. Its funding post names companies including Google, Microsoft, Slack and Databricks, and says these partnerships take Replit from a place to prototype ideas to a platform where software can be deployed to support real teams, power real workflows and serve real customers at scale. Chroma’s context technology now becomes part of that platform, available to the same enterprise customers.

ChromaDB is the reason Chroma is so well known among AI developers. Huber writes that after its release in early 2023 it exploded in popularity, and that at one point Chroma accounted for over 95% of the downloads for vector search. Vector search is the technique that lets software find information by meaning rather than by exact keywords, which is why it sits at the heart of most AI applications that answer questions about a company’s own documents.

Growth came fast. Chroma’s seed announcement noted that in the five weeks after launch the project crossed 35k Python downloads in a single month. Today the announcement puts open-source Chroma at 6M downloads a month and just shy of 30k GitHub stars. The company says it invested deeply for years to turn ChromaDB into an open-source, distributed and serverless database, and that it “remains unparalleled for its speed, cost-profile, multi-tenancy and scalability”.

The customer list shows how far the database has travelled from developer side projects. Chroma says ChromaDB powers workloads for customers including SpaceXAI, Mintlify, Krea, Retool and Coreweave, and tens of thousands of others. Chroma’s website describes the product as fast, serverless and scalable infrastructure supporting vector, full-text, regex and metadata search, built on object storage and licensed under Apache 2.0, with SOC 2 Type II compliance for business customers.

Context Rot and Context-1 research stands the test of time

Chroma has also built a strong reputation for research. Huber writes that the company’s technical reports on Context Rot, published in July 2025, and Context-1, published in March 2026, “have stood the test of time”. The team has also published engineering deep dives on Fission transactions, object-storage native retrieval, wal3, its write-ahead log on S3, and its query execution engine. For Replit, that research base arrives together with the people who produced it.

Foundation is the newest part of what Replit is acquiring, and the part most closely tied to the deal’s stated goal. In the middle of 2026, Huber writes, it became apparent that “the next inflection point in our growth was going to come from serving enterprise context”, which Chroma calls the context layer. The company began work on Foundation as its answer: self-improving context for organisations.

Chroma’s product page describes how it works. Foundation learns from agent sessions and maintains a self-improving wiki of everything a person and their team know. It follows along with a team’s work in tools such as Claude Code, Codex, Cursor and Slack, finds both explicit and tacit knowledge about how work gets done, and turns it into pages that agents can read through MCP, a command-line tool or an API. Connectors for Notion, GitHub, Google Drive and Granola are listed as coming soon.

The results came quickly. According to the announcement, Foundation hit state of the art on the BEAM memory benchmark and trended on Ramp insights immediately after release. The product page adds that Foundation is built on ChromaDB and Context-1, Chroma’s retrieval engine, and ships with versioning and diffs, lineage and citations, concurrency control, real-time sync and continuous background improvement that incorporates human feedback.

SOC 2 Type 2 and $30 per user from day one

Foundation was designed for enterprise buyers. Chroma lists SOC 2 Type 2 certification, granular access control, SSO and SAML, bring-your-own-cloud deployment, dedicated support and a service-level agreement. Coding agent sessions are visible only to their author. Pricing is set at $30 per user per month plus overages, billed as infrastructure across database operations, embedding inference and LLM inference.

This is where the fit with Replit is clearest. Replit’s March 2026 announcement says users from 85% of the Fortune 500 build with the platform. Foundation gives those enterprise teams a way for Replit’s agents to know what their organisation knows, and Replit gives Foundation a direct route to those teams through its scaled go-to-market engine. Chroma’s announcement says Replit will extend the open-source project for enterprise context, which places Foundation’s ideas at the centre of Replit’s enterprise offer.

The most immediate change applies to Chroma Cloud, the company’s hosted service. The announcement states that over the next 60 days Chroma is “responsibly winding down our existing cloud service”, and a separate wind-down page published the same day sets out two milestones. The schedule gives customers clear dates and a clear checklist, which is the most helpful thing a cloud provider can offer during a transition.

On 22 November 2026, thirty days after the notice, Chroma Cloud becomes read-only. Applications can no longer create, update or delete data there. Before that date, Chroma asks customers to stop or redirect workloads that write to Chroma Cloud, move active workloads to another Chroma deployment or database, and begin exporting and validating any data they need to retain. Existing data remains available for reads during the following 30-day read-only period.

On 22 December 2026, sixty days after the notice, Chroma Cloud stops serving read requests. By then customers should have finished exporting data, verified that migrated applications no longer depend on Chroma Cloud endpoints, and removed or rotated Chroma Cloud credentials stored in their applications and infrastructure. Chroma directs customers who need help with their account to its support team.

Private code moves into the Apache 2.0 repository

The open-source project moves in the opposite direction: it is getting stronger. Chroma says it has committed additional code from its private repositories to the Apache 2.0 repository “to give the strongest continuity story for open-source Chroma”. The wind-down page confirms that the transition applies only to the existing Chroma Cloud service, and that open-source ChromaDB remains available under the Apache 2.0 licence.

Replit has also committed to the project’s future. In the announcement’s words, “Replit will continue to contribute back to the project as it extends it for enterprise context”. For the tens of thousands of teams running ChromaDB themselves, that means the database they depend on keeps its permissive licence, gains code that was previously private, and now has the backing of a company valued at $9 billion.

Context joins models at the core of the agent stack

For founders building AI infrastructure, the deal shows how deep open-source adoption, a credible research record and a clear enterprise product come together into a compelling strategic fit for a fast-growing platform. Chroma’s path ran from a developer tool in 2023 to enterprise context in 2026, and each step made the company more valuable to a buyer whose own agents need memory. For boards at AI application companies, the deal is a clear signal that the data and context layer is becoming as central to agent products as the models themselves.