Onyx Review (2026): Pricing, Features & Honest Verdict

TLDR

Onyx is an open source AI chat and enterprise search platform that connects to your company’s docs, apps and tools. We found it the most credible self-hosted answer to Glean, and the price gap is enormous. Best for: engineering-led teams that want internal AI search without a six figure contract. Price: free self-hosted, cloud from $20/user/mo. Rating: 7.7/10

What is Onyx?

Onyx is an open source AI assistant that sits on top of your company’s knowledge. You point it at Slack, Google Drive, Confluence, GitHub, Salesforce and dozens of other sources, it indexes them, and then people ask questions in plain language and get answers with citations back to the original documents. Under the hood it runs hybrid search, contextual retrieval and an LLM-built knowledge graph rather than naive vector lookup. On top of search it adds custom agents, actions through MCP and OpenAPI, web search, a code interpreter, image generation and a Slack bot so answers reach people where they already work.

The project started life as Danswer, founded in 2023 by Yuhong Sun and Chris Weaver, went through Y Combinator, and rebranded to Onyx. In March 2025 the company raised a $10M seed round co-led by Khosla Ventures and First Round Capital with Y Combinator participating. The repository has passed 20,000 GitHub stars. The reason it matters now is the shape of the market: Glean and its peers are excellent and sold on six figure enterprise contracts with seat minimums. Onyx offers the same category of product as an MIT-licensed download you can run on your own infrastructure, which changes who can afford internal AI search at all.

What Are Onyx’s Key Features?

Connectors and Permission Inheritance

Onyx ships more than 40 plug and play connectors covering the usual enterprise surface: Slack, Google Drive, Confluence, Jira, GitHub, Notion, Salesforce, Zendesk and the rest. They sync on a schedule so the index stays current. The part that actually matters for a company is permission inheritance. If a document is restricted in Google Drive, Onyx respects that when answering, so a contractor asking a question does not get a summary of the board deck. This is the feature that separates a real internal search tool from a weekend RAG project, and it is present in the open source edition rather than paywalled.

Retrieval Built for Scale

The retrieval stack combines keyword and vector search with contextual retrieval and a knowledge graph layer, and Onyx states it stays performant and accurate into the tens of millions of documents. The company published a benchmark of 99 workplace questions across 220,000 internal documents where it reports win rates of 64 percent against ChatGPT, 68.1 percent against Claude and 76 percent against Notion AI. Vendor benchmarks are vendor benchmarks and should be read with the usual scepticism, but the underlying claim is reasonable: a model with indexed access to your actual documents will beat a general assistant that has never seen them.

Bring Your Own Model and Self-Hosting

Onyx is LLM agnostic. You can point it at OpenAI, Anthropic, Google or an open weights model running locally, and swap providers without changing the rest of the stack. Deployment is documented for Docker, Kubernetes and Terraform with guides for the major clouds. For teams under data residency rules, or anyone who simply does not want their internal wiki leaving their VPC, this is the whole pitch. It also means your ongoing cost is infrastructure plus model tokens rather than a per seat licence.

Agents, Actions and Deep Research

Beyond question answering, Onyx supports custom agents with their own instructions and document scopes, actions via MCP and OpenAPI so an agent can call your internal APIs, a deep research mode for multi step investigation, a code interpreter and web search. Admins get query history and usage dashboards, which is the practical way to find out what people cannot find, and that turns into a documentation backlog. The Slack integration is the highest leverage piece for most companies, because internal search only pays off when it lives where questions get asked.

How Much Does Onyx Cost?

Onyx starts at zero. The Community Edition is MIT licensed and free to self-host with unlimited users, covering chat, RAG, agents, actions and the connector library. You pay for servers and model tokens, nothing else. That is a genuinely unusual position in enterprise search, where the norm is a call with sales before you see a number.

If you would rather not run it, Onyx Cloud’s Business plan is $20 per user per month on annual billing. That includes the chat and search interface, access to major LLMs, custom agents, actions, the 40 plus connectors, web search, deep research, code interpreter, image generation, the Slack integration, developer APIs, query history, usage dashboards, Google OAuth, role based access control and permission inheritance. Enterprise is custom priced and adds OIDC and SAML SSO, white labelling, on premise and region specific deployments, custom integrations, data exports, invoice billing, volume discounts, a dedicated support contact and an SLA. Onyx is SOC 2 Type II certified and GDPR compliant.

Set that against the field. Glean does not publish pricing at all, and buyer reports put it in the range of $45 to $75 per user per month with roughly 100 seat minimums, meaning annual contracts starting near $60,000 and large deployments running far higher. Dust publishes 29 euros per user per month for Pro with enterprise pricing above 100 members. So for a 100 person company, Onyx Cloud lands around $24,000 a year, Dust sits higher, and Glean is a different order of magnitude. Self-host and the licence cost goes to zero, though you are now paying an engineer to keep it healthy, which is a real cost people routinely forget to count.

Who is Onyx Best For?

Use Onyx if you have engineering capacity and knowledge spread across Slack, Drive, Confluence and a dozen SaaS tools that nobody can find anything in. Use it if data residency or sovereignty rules mean your documents cannot go to a third party index, because self-hosting with your own model endpoint solves that cleanly. Use it if you have priced Glean, felt the seat minimum, and want the same capability at a fraction of the number. Use it if you want to own the system: full code visibility, your own model choice, no vendor deciding what your retrieval pipeline does next.

• Engineering-led SaaS companies of 30 to 500 people with sprawling internal docs.
• Regulated or defence-adjacent teams that need on-premise deployment.
• Agencies and consultancies that want per client document scopes and their own branding.
• Teams already running their own model infrastructure who want a retrieval layer on top.

Skip Onyx if nobody on your team wants to own a Kubernetes deployment and you are unwilling to pay for the cloud version. Skip it if you need the polished, hand-held enterprise rollout that Glean’s sales and customer success motion provides, because open source support is community support until you sign an enterprise contract. Skip it if your entire knowledge base already lives in one tool, since Notion AI or Google’s own search will cover you for less effort. And be realistic about the setup: connectors, permission mapping and index tuning take real time before the answers get good.

Best Onyx Alternatives

Glean

The category leader and the benchmark Onyx is explicitly measured against. Glean has broader connector coverage, more mature governance and an enterprise motion that will hold your hand through rollout. It does not publish pricing. Reported deals sit in the $45 to $75 per user per month range with around 100 seat minimums, putting entry-level annual contracts near $60,000 and large deployments well past $200,000. Choose Glean if you are a large enterprise with the budget and you want the safest procurement decision rather than the cheapest one.

Dust

Dust is the managed middle ground: transparent pricing at 29 euros per user per month for Pro with a 14 day trial from a single user, access to frontier models, and custom assistants built on your company data. Enterprise pricing starts from 100 members with SSO, SCIM and a choice of US or EU hosting. It is faster to stand up than self-hosted Onyx and cheaper than Glean, but you do not get the source code or the option to run it in your own VPC on the standard plan. Choose Dust if you want an assistant platform today and self-hosting is not a requirement.

Open WebUI

The free, self-hosted option if your real need is a good chat interface over your own models rather than full enterprise search. It is genuinely excellent at what it does and costs nothing to run, but it is a chat front end with document support rather than a connector-driven index with permission inheritance across 40 SaaS tools. Choose Open WebUI if you mostly want private chat with local models and a handful of uploaded documents. Choose Onyx once the questions people ask depend on knowing what is in Slack and Confluence.

Final Verdict: Is Onyx Worth It?

Onyx earns a 7.7 because it does something the rest of this category refuses to do: it publishes a price and gives you the code. The retrieval quality is serious, permission inheritance is handled properly rather than promised for later, the connector library is broad, and $20 per user per month on cloud is roughly a third of what buyers report paying for Glean. The MIT-licensed self-hosted edition with unlimited users is the strongest offer available to any team with an infrastructure engineer.

What holds it back is the same thing that makes it good. Self-hosting is work, community support is community support, and the polished enterprise rollout experience lives behind a custom contract. Vendor-run benchmarks should not be your buying criteria either. But if you have looked at internal AI search and walked away because of the price, Onyx is the one to try. Run the cloud trial to see whether the answers are good on your data, then decide whether to bring it in house.