Chroma AI Database for Private Business Search
Chroma is an open-source AI database that MILabs helps businesses set up for private search over their own documents.
Updated 2026-02-27
What Chroma is
Chroma is an open-source database built for AI applications. It stores text as embeddings, which are numerical representations of meaning, alongside the original documents and their metadata. When a question arrives, the database finds the stored passages closest in meaning to it, rather than only those that share the same words. Developers reach it through Python and JavaScript clients, and it can run inside an application or as a separate server the business controls. Chroma is one component in a larger AI system and does not write answers on its own. A language model or another tool usually sits on top of it, reads the passages it returns and produces a reply. The project publishes its own documentation at trychroma.com, which is the reference for what the software does and how it is run. For a business, the value lies in searching internal material by meaning while the source files stay where the business already governs them.
Chroma in plain language
Searching by meaning depends on a preparation step that happens before anyone asks a question. Long documents are first split into smaller passages, since a few paragraphs can be more precise than a whole manual. Each passage is then turned into an embedding by an embedding model, and that choice matters because vectors from one model cannot be compared with vectors from another. Change the model later and the whole library must be processed again.
Each passage is stored with labels recording its origin, such as the file name, the section heading and the date it was last revised. Those labels let any result be traced to its source. They also let a search narrow to one client, one year or one document type before the closest matches are chosen.
Retrieval is only one part of an answer. The passages returned are placed beside the question, and a language model is asked to respond using only that material. Quality therefore depends on chunking, the embedding choice and the wording of the instructions as much as on the database.
What businesses use it for
- Helping staff find the right section of a policy, procedure or handbook by describing what they need
- Grounding an assistant's answers in the company's own product sheets, contracts or case files
- Matching incoming enquiries to past questions and approved answers
- Keeping a searchable record of client correspondence and meeting notes for account teams
- Supporting compliance reviews by retrieving the clauses that apply to a given matter
Who it suits
- Firms with large document libraries that staff search every day
- Teams that want an AI assistant to answer from the business's own material
- Owners who need the data kept on infrastructure they control
How MILabs sets it up
The first conversation with the owner or team lead covers the documents and records that hold the knowledge people need. From there we choose a setup that fits, on a machine the business already manages or on agreed infrastructure. We connect the sources the business already uses, load the material and attach the metadata that keeps every result traceable. Access is set by role so staff see only what their work requires, and permissions are reviewed as people change roles. Before anything goes live, we agree in writing who looks after the database, who updates the source material and who responds when a result looks wrong. Outside Sydney, this work is delivered remotely.
Builds it fits
Industries it fits
Chroma in practice
In a working office, the database sits behind a single question box or a chat window that staff already use. A person types what they need, and the system returns a short list of passages with their source labels. The person opens the original document to confirm the wording before acting on it, because the passages are a starting point for judgement. That is why the source link stays visible in every result.
Each body of material has a named person who decides which version is current, adds newly approved documents and removes withdrawn ones. When a policy is replaced, the old passages must come out of the store as well as the new ones going in, otherwise the system will confidently quote a clause nobody uses any more.
Staff feedback becomes the maintenance log. When a result looks wrong, the reviewer traces it back to the passage and its label, then decides whether the fix belongs in the source document, in how the material was split or in the instructions given to the assistant. Over time that record shows which parts of the library need tidying most.
Frequently asked questions
- Which builds work well with Chroma?
- Chroma suits builds that find relevant passages in a body of text before answering. The staff knowledge assistant, the document question-and-answer assistant and the client FAQ assistant all depend on that step. The knowledge base chatbot and the private document server follow the same pattern for different audiences. Policy compliance and proposal drafting builds gain from it too, since the right clause must be found first.
- What is Chroma?
- Chroma is an open-source database for AI applications. It stores embeddings, which are numerical representations of text meaning, together with the source passages and their metadata. A query is converted into the same form, and the database returns the stored items closest to it. Chroma does not generate answers itself, so developers usually pair it with a language model. The project's documentation at trychroma.com sets out its current features.
- How does MILabs set up Chroma?
- MILabs begins with a conversation with the owner or team lead about where the business's knowledge lives. From there we agree whether Chroma runs on hardware the business already manages or on infrastructure chosen with it. We connect the sources, load the material and set access by role. Ownership of updates and of result reviews is written down before go-live. Outside Sydney the work is delivered remotely.
- Where does Chroma store data?
- When Chroma is self-hosted, data is written to the storage location configured for that installation, on the machine or server where it runs. The business should choose that location carefully. MILabs records where the files sit, who can reach them and how they are backed up. Source documents usually stay in the systems they came from, and Chroma holds the searchable copy alongside them.
- When is Chroma the wrong choice?
- Chroma is a poor fit for a few documents that a shared folder already handles well. It also fits poorly when the team cannot agree who maintains the source material, because results are only as current as what is loaded. Where a company already runs a mature search platform, another store may only add upkeep. In those cases MILabs will say so during the first conversation.
