Glean cost vs self-hosted Onyx: where the money moves
Imagine a CIO receiving a Glean renewal that combines per-user Core Suite licensing with a Model Hub usage commitment, and a CFO asking whether running Onyx would be cheaper. The honest answer is that self-hosting Onyx replaces a per-user subscription and usage commitments with infrastructure, implementation, connector work and operating staff. It is cheaper only when the user count and query volume justify those fixed costs. The Onyx license carries no fee: Onyx Community Edition is MIT licensed, and the Enterprise Edition is a separate paid option you would need to price on its own. No Glean price, and no DeployOpen price, appears on this page. Glean's Core Suite and Model Hub figures must come from a written quote from Glean, and you should ask for every line item in writing before comparing.
The cost split is different on each side. Glean's subscription recurs for each renewal term, and its usage commitment recurs as well. Self-hosting shifts most of the spend into one-time work, including the initial deployment, connector configuration and permission mapping for each source, the migration of agents and search settings, and a pilot group. Recurring costs then include the compute that runs the index, database, Redis and model servers, any inference capacity, storage, backups, upgrades, monitoring, and the staff or support contract that keeps the stack patched. Model inference is often the largest variable: a hosted model billed per token scales with use, while a self-hosted model such as one served with vLLM converts that spend into accelerator capacity you reserve whether or not it is busy.
Before you compare totals, confirm four things. Ask Glean to itemize Core Suite, Model Hub and any Customer Hosted fees, and to state the term and true-up rules. Estimate your own connector count, index size and peak concurrent users. Price the GPU or other accelerator capacity you would need for your model choice, and get a number from your cloud or hardware provider rather than assuming one. Finally, name who will operate the stack, whether internal engineers or an implementation partner under a fixed-scope agreement, because that staffing decision often moves the answer more than licensing does. A simple model with these inputs, built on your own numbers, shows where the break-even point sits for your organization.
Which cost components sit on each side
This table lines up each cost component of a Glean contract against the matching cost on a self-hosted Onyx deployment, and marks whether it is one-time or recurring. Read the Glean column against your order form and the Onyx column against your own infrastructure and staffing plan. The main takeaway is that Onyx Community Edition removes the per-user license line, but the model, index, database, connector and operations costs move onto your budget and must be estimated from your own environment.
| Cost component | Glean side | Self-hosted Onyx side | One-time or recurring |
|---|---|---|---|
| Software license or subscription | Per-user Core Suite licensing, confirm seat counts and renewal terms in the order form | Onyx Community Edition is MIT licensed with no license fee; Enterprise Edition only if you need premium features | Glean: recurring. Onyx: none for Community Edition |
| Model inference | Model Hub usage commitments, confirm what is included and what is metered | GPU or accelerator capacity for vLLM or other model servers, or fees to a proprietary provider you choose | Recurring |
| Index and database hosting | Included in the Glean service, confirm where indexes and logs are stored | Postgres, Vespa or OpenSearch, Redis and object storage run in your cloud account or data center | Recurring |
| Connector build and maintenance | Glean's connector catalog, confirm coverage for each source you need | Community connectors exist for SharePoint, Confluence, Jira, Slack and others; you own changes when source APIs change | Build is one-time; upkeep is recurring |
| Implementation and migration | Onboarding and admin setup, confirm what the contract includes | Fixed-scope planning, deployment and integration by your team or a partner such as DeployOpen, priced from written scope | One-time |
| Enterprise Edition or premium features | Included in the subscription, confirm which features such as audit and admin controls apply | Enterprise Edition license for features such as RBAC, if required; confirm terms directly with Onyx | Recurring if licensed |
| Upgrades and patching | Glean-managed in both Hosted and Customer Hosted models, confirm the upgrade notice terms | Your team applies releases and Helm chart updates on your own schedule | Recurring |
| Internal skills and operations staff | Smaller admin load for your team, confirm which tasks remain yours | Engineers for Kubernetes or Docker, model serving, monitoring and incident response | Recurring |
| Support | Vendor support under the Glean contract, confirm response targets in writing | Community channels, or a paid retainer with a partner for ongoing support and upgrades | Recurring if contracted |
What moves the comparison most
Glean's cost is mostly set by seats and usage commitments, while a self-hosted Onyx stack shifts spend toward compute, storage, connector work and people. The six drivers below are where the two sides diverge most, so measure them before comparing invoices.
Active users and seat tiers
Glean's per-user Core Suite licensing scales with the people who log in, so the cost follows headcount and any tiering in the order form. Onyx's Community Edition has no per-seat fee, but your compute, storage and support load still grow as more people query the system. Measure monthly active users by department, not total licensed accounts, because idle seats are where renewals usually carry the most waste.
Indexed volume and source count
Glean's pricing and your Onyx costs both depend on how much content is indexed and how many systems are connected. More sources mean more connector sync jobs, more vector and keyword index storage, and more permission-sync work on your side. Ask Glean how its commitment treats documents and sources, and inventory your own document count, total size and connector list before estimating index hosting.
Model choice and token volume
Glean's Model Hub usage commitments price model access by consumption, so heavy assistant use raises spend without any change in seat count. With Onyx, the model is your choice: a hosted proprietary provider bills per token, while self-hosted models through vLLM or Ollama turn that cost into GPU capacity and operations. Measure queries per day and average tokens per answer on a pilot group, since this is the input most sensitive to usage.
Availability and on-call needs
Glean Hosted and Customer Hosted both place infrastructure upgrades, patching and monitoring with Glean. A self-hosted Onyx stack moves those tasks, plus after-hours response for the index, database, Redis and model servers, onto your team or partner. Define the required availability window and response expectations in writing, then price the on-call rota that would cover them, because this cost is often missing from self-hosted estimates.
Edition choice
Onyx comes as Community Edition under MIT, as a self-hosted Enterprise Edition with features such as RBAC, or as Onyx Cloud. The edition decision changes the licence line item, the operational scope and which compliance controls you can run yourself. Confirm which features your audit requires, such as SCIM provisioning or query history auditing, and get the Enterprise Edition quote separately from infrastructure estimates.
In-house Kubernetes and GPU skills
Running Onyx on Kubernetes or Helm, and vLLM on GPUs with tensor or pipeline parallelism, requires skills that many platform teams do not have in-house. Staff time, training or an implementation partner becomes a one-time cost, and ongoing upgrades become a recurring one. Assess your current cluster and GPU operating experience honestly, because a skills gap moves the comparison more than most line items.
How the Onyx footprint changes the bill
Onyx Lite is the lighter deployment for chat and agents without a vector database or background workers. Because the index tier and sync workers are absent, the infrastructure bill is smaller, but the platform also has less to work with on company content. Confirm with Onyx which connectors and search features remain available in this mode before you size anything.
- Application and database lines: The application services and a relational database remain, with no vector index cluster to size, patch or back up.
- Model usage: Inference is billed by the provider or self-hosted model server you connect, so token or accelerator cost depends on your usage rather than on the Onyx license.
- Staff tasks: Your team still handles container updates, TLS certificates, secrets, chat history backups and service monitoring.
- Quote request: Ask for the list of required services, the supported model providers and whether connector sync is included.
Check whether the search use case in your Glean contract depends on indexed content, because Lite may not cover it.
Onyx Standard is the configuration that covers enterprise search, with vector and keyword indexing, background sync workers, model inference servers, object storage and Redis. Its documented infrastructure includes Postgres, Vespa or OpenSearch for the index, and Redis, so each becomes a line you provision, size and patch. Model serving is often the largest variable. vLLM is one open-source option for it, and it has no paid edition, so its cost is accelerator capacity, the surrounding infrastructure, implementation and the people who operate it.
- Index and database hosting: Postgres plus Vespa or OpenSearch need storage, memory and backup schedules that grow with the number of connected sources.
- Sync workers and cache: Background workers and Redis run continuously to keep documents and permissions current.
- Object storage: Holds files and artifacts produced by sync and indexing jobs, and it must be included in backup scope.
- Model inference: vLLM exposes OpenAI-compatible endpoints and runs on NVIDIA CUDA, AMD ROCm and Intel XPU GPUs, so accelerator capacity is the main recurring cost.
- Operator time: Upgrades, index tuning, incident response and connector changes after source application updates are staff tasks.
Ask for a quote that prices each service separately, states the GPU sizing assumptions and estimates operator hours per month.
Onyx is available as self-hosted open source, as Onyx Cloud, a managed service, and as a self-hosted Enterprise Edition with premium features such as role-based access control. In the managed option, Onyx operates the infrastructure, so most infrastructure lines move into the vendor's fee, and your team keeps configuration and user administration. The Enterprise Edition is a separate line item. Code in the ee directories is under the Onyx Enterprise License, so it is priced apart from the Community Edition and from the hardware you run.
- Onyx Cloud subscription: A recurring fee for the managed service. Confirm what it covers, including model usage, storage limits and data location.
- Enterprise Edition license: A separate recurring license for premium features. The README lists SSO, SCIM provisioning and query history auditing among enterprise features, so confirm which ones are included.
- Retained staff tasks: Identity configuration, connector permissions and access reviews stay with your team even in the managed option.
- Exit and data terms: Ask how indexed data, configuration and query history can be exported if you move off Onyx Cloud later.
Request one written quote that separates the managed fee, the Enterprise Edition license and any self-hosted infrastructure you keep.
How to build the Glean vs Onyx model with your own numbers
This sequence turns a Glean renewal into a like-for-like model, with your seat count, source volumes and model choices on one side and the Onyx footprint, staffing and migration costs on the other. The result is a three-year view you can defend in a budget review.
Collect your current Glean contract terms
Start with the renewal notice and the signed order form. Record the number of licensed seats, the Core Suite terms, the Model Hub usage commitment, the contract term and renewal date, and any fees tied to Customer Hosted deployment. Confirm with your Glean account team which lines are fixed and which can change during the term. The step is done when every line on the renewal notice is matched to a field in a single assumptions sheet, with the source document named beside it.
Request an itemized Glean quote
Ask Glean for a quote that separates Core Suite seats, Model Hub commitments and any Customer Hosted fees, so each cost line can be compared with an Onyx component later. Ask how usage above the commitment is billed, how seat counts change mid-term, and what happens to committed spend at renewal. Ask whether the quote covers the deployment model you run today. The step is done when you hold a written quote with each component priced separately and the billing rules for overage are stated.
Inventory every source system and its volume
List each system Glean searches today, such as file shares, collaboration suites, ticketing, chat and CRM, along with the owner, the approximate document or message volume, and how permissions are set in each one. Mark which connectors Onyx supports and which would need custom work. Note where permissions are inherited from groups, because permission sync is part of the cost of matching Glean's behavior. The step is done when every source has an owner, a volume estimate and a connector status, so connector effort can be scoped.
Choose a model strategy and measure tokens in a pilot
Decide which models answer queries and which self-hosted options you will test, such as Ollama, LiteLLM or vLLM for open-weight models, or a proprietary provider. Run a pilot group on real queries for a fixed period you set, and record prompts, answers, and input and output tokens per user per month. Project monthly tokens from that data rather than from vendor marketing. If you use self-hosted inference, note that vLLM has no paid edition, so its cost sits in accelerator capacity, surrounding infrastructure, implementation and operations. The step is done when you have a measured monthly token figure and a stated model choice with its hardware assumptions.
Price the Onyx footprint with a fixed-scope quote
Choose between Onyx Lite and Onyx Standard based on whether you need vector and keyword indexing and background sync workers. Size the database, index, Redis, model servers and object storage for your document volume and user count, and decide whether you need the Enterprise Edition for features such as RBAC or SSO and SCIM provisioning. Request a fixed-scope implementation quote from DeployOpen or another partner, covering connectors, permission mapping, hardening, backup and documentation. Ask for the cloud or data center costs separately from the labor. The step is done when infrastructure, licensing and implementation are each listed as their own lines.
Add staff time for operations and support
Estimate the people who will run the platform: patching and upgrades, connector maintenance as source APIs change, monitoring, on-call coverage, and user support. For Glean, the vendor carries these tasks in both hosted and Customer Hosted models, so the staff line is the one most often missing from a comparison. Decide whether this work sits with existing platform engineers, a new hire or a retainer with a partner, and record the share of time each role will spend. The step is done when each operations task has a named owner and an hours-per-month assumption you can defend.
Build the three-year cost view
Create a three-year table with two sections: one-time migration costs, including connector build, permission mapping, pilot runs and parallel running of both systems, and recurring costs, including infrastructure, Glean or Onyx subscriptions, model usage and staff time. Apply the same usage and seat assumptions to both sides, and show the sensitivity to each assumption. Record every assumption in the sheet with its source, so the model can be reviewed and changed. The step is done when the total for each year and the one-time subtotal can be traced back to a quote, a measurement or a named assumption.
What the migration work costs in effort
The one-time cost of moving from Glean to self-hosted Onyx sits mostly in connectors and permissions, not in the software. Work through each Glean data source in turn and confirm that Onyx has a connector for it. The Onyx connectors directory lists SharePoint, OneDrive, Outlook, Microsoft Teams, Google Drive, Gmail, Confluence, Jira, Slack and Salesforce among others, but a listed connector is not proof that your tenant, scopes and content types are covered. Build a source-by-source matrix that records the connector, the authentication method, the sync schedule, and any content Glean indexes that Onyx will not. That matrix becomes the basis for your fixed-scope agreement.
Permissions are the item most often underestimated. Onyx syncs source permissions so that answers only draw on documents a user can already see in the source application. That is the control you want, but it means access testing belongs in the migration budget. Plan test accounts for each role and group, run known-restricted queries as each one, and compare results with what the source system shows. Recreating agents is a separate task. Each Glean agent or saved configuration needs its prompts, scoped sources, and model choice rebuilt in Onyx and reviewed by the team that owns it. In the quote, ask for deliverables and acceptance criteria rather than hours. Useful items include a connector and permission matrix signed off by your security team, a pilot group with a defined query set and pass criteria for search quality and access, a cutover plan dated against your Glean renewal, and a parallel-running period in which both systems answer the same questions. Ask who runs each acceptance test, what happens if a connector fails its permission check, and how the written scope handles sources added during the pilot. Confirm the renewal and notice dates with Glean directly, since they set the length of parallel running you can afford.
Questions buyers ask before comparing Glean and Onyx costs
What does the Onyx license cost compared with Glean?
Onyx Community Edition is MIT licensed, so the software itself carries no license fee. The Enterprise Edition code in the ee directories is under the Onyx Enterprise License, so it needs a separate commercial agreement. Glean does not publish a price list, and its per-user Core Suite licensing and Model Hub usage commitments come from a written quote. Compare the two by putting Glean's quoted subscription next to the infrastructure, model, connector and staffing costs of Onyx, and ask Glean exactly which features and usage volumes the quote covers.
How long does a migration from Glean to Onyx take?
The duration depends on the drivers you control: how many connectors you must rebuild, how many source systems keep their permissions in sync, how many agents need recreating, how much model selection and testing you need, and how much data you index. Your team's familiarity with Kubernetes, Postgres and identity systems also matters. Running a pilot group on real queries before the Glean renewal date gives you measured results to plan around, so ask any implementation partner to base its timeline on a written scope rather than a generic estimate.
Can our team run Onyx without an outside partner?
Yes, if your team has the skills to operate the stack. Onyx Standard uses Postgres, Vespa or OpenSearch for the index, Redis, model servers and object storage, and it deploys with Docker, Kubernetes, Helm or Terraform. You will need people comfortable with Kubernetes operations, Postgres backup and upgrades, GPU scheduling and driver management if you host models yourself, and identity integration through Google OAuth, OIDC or SAML with SCIM provisioning. Unlike Glean, which patches and upgrades Customer Hosted deployments, you are responsible for patching and upgrades on your own schedule.
Do we need the Enterprise Edition for RBAC and SCIM?
Onyx lists role-based access control and SCIM user provisioning among its enterprise features, and these sit in the Enterprise Edition under the Onyx Enterprise License. The Community Edition covers RAG, AI chat, agents and actions, and it supports SSO through Google OAuth, OIDC or SAML. Onyx also offers a managed Onyx Cloud service, which is a different option from the self-hosted Enterprise Edition. Confirm with Onyx which features are in the Community Edition for your version, and price the Enterprise license as a separate line in your comparison.
How does running vLLM compare with paying for hosted models?
vLLM has no paid edition, so its cost is the accelerator capacity, the surrounding infrastructure, implementation and the people who operate it. Hosted model usage is billed by consumption under Glean's Model Hub commitments. The break-even point depends on your query volume, the model size you need, GPU utilization and how many hours the hardware sits idle. vLLM's OpenAI-compatible endpoints let Onyx use a self-hosted model without changing the application, so run a load test on your own prompts and ask Glean and your hardware provider for pricing based on those measurements.
How do we keep Onyx upgrade effort low?
Keep the deployment close to the upstream Helm chart or Docker configuration, record every change in version control, and pin versions so upgrades happen on purpose. Onyx v4.9.2 was released on October 15, 2026, according to its GitHub releases page, and the project publishes a Helm chart, version 0.9.17 as of October 16, 2026. Read release notes before each upgrade, test in a staging environment that mirrors production, and schedule regular upgrade windows. Automating backups of Postgres and the index before each change reduces the risk of rolling back.
What should we ask a partner for in a quote?
Ask for a written scope that lists each deliverable: the connectors and permission sync per source, the model choice and hosting plan, the infrastructure design, identity integration, backup and recovery tests, documentation, and handover training. Ask which items are one-time and which recur, such as upgrades, monitoring and support, and what drives each cost up or down. Request assumptions about data volume, user count and connector count, and ask who is responsible for patching and incident response after go-live. DeployOpen provides a quote from a written scope and does not publish a price list.
Can DeployOpen take over Onyx operations under a retainer?
DeployOpen offers ongoing support, upgrades and continuous testing as retainers, so it can operate an Onyx deployment after the initial implementation. Scope the retainer in writing before work starts, covering which systems are included, how upgrades and backups are handled, response expectations and what remains your team's responsibility. DeployOpen deploys inside your own cloud accounts, data centers or air-gapped networks, so the environment and data stay under your control. A retainer quote follows from that written scope.
How DeployOpen scopes a fixed-price Onyx implementation
DeployOpen plans, implements, integrates, hardens, documents and supports self-hosted open-source platforms inside your own cloud accounts, data centers or air-gapped networks. For a Glean cost comparison, that means turning your connector list, permission model and model choices into a written scope before any build starts.
How we scope
- Inventory the Glean connectors, sources and permission rules your users depend on today.
- Agree which Onyx Community Edition or Enterprise Edition features the scope needs, such as SCIM or RBAC.
- Document the inference, index, database and staffing assumptions so the quote follows from the written scope.
How we build
- Deploy Onyx Standard or Onyx Lite with Docker, Kubernetes, Helm or Terraform in your environment.
- Connect the agreed sources and verify that synced permissions match each source application.
- Pair the deployment with the model server you choose, such as Ollama, LiteLLM or vLLM, and test query quality on real questions.
How we hand over
- Deliver architecture, runbooks and configuration documentation covering indexing, upgrades and backups.
- Train the operators who will patch, upgrade and monitor the stack after handover.
- Provide evidence of access controls and logging that your compliance team can review with its own auditor.
How we support
- Offer ongoing support, upgrades and continuous testing as a retainer, separate from the fixed-scope build.
- Track upstream Onyx and Helm chart releases so your team can plan upgrade windows.
- Confirm with DeployOpen which retainer terms, response expectations and deliverables apply before signing.
