Which open-source alternative to ChatGPT Enterprise fits your situation
If you are facing a renewal, a data residency requirement or a pricing change, the fit depends on what you must keep under your control. Teams that need chat with data kept in their own cloud account or data center usually look at Open WebUI, LibreChat or AnythingLLM paired with a self-hosted model or an API provider you contract with directly. Teams whose main need is answers grounded in company documents and connectors to business systems should look at Onyx, which is covered in the sibling pages on Microsoft 365 Copilot and Glean.
The reason teams look for alternatives is structural. According to OpenAI's documentation, ChatGPT Enterprise runs only as an OpenAI-hosted service. The controls OpenAI documents for it are SAML SSO, admin retention settings, regional data residency and SOC 2 Type 2 audits. Those controls are useful, but none of them let you host the application, choose where inference runs at the level of your own network, or hold the chat stack in the same change process as the rest of your infrastructure. If your policy requires customer-hosted deployment, or if you want to change model providers without changing the interface your staff use, the hosted model is the constraint to test first.
Before you shortlist a platform, write down three things: which identity provider must sign users in, how long conversations must be retained and in which region, and whether you need an export path that your compliance team can run without the vendor. Then check each candidate against those requirements. Confirm SSO and provisioning options, retention and deletion behavior, and audit logging directly in the project documentation and in your own test environment, since these details change between releases. The answer will often be a hybrid: a self-hosted interface for regulated workloads and a hosted model for lower-risk use.
Open WebUI, LibreChat, AnythingLLM and Onyx compared
These four platforms cover the chat interface that ChatGPT Enterprise users see, but they differ in how they connect to models, how they handle identity, and how much of your company content they can search. Compare them on licensing, authentication, and whether you need chat only or chat grounded in internal documents.
Open WebUI
A self-hosted AI interface that works with Ollama and OpenAI-compatible APIs, so it can run fully offline when paired with local models. It installs with Docker, Python pip or on Kubernetes. The license is the Open WebUI License, a BSD-3-Clause style license with a clause that restricts removing or altering Open WebUI branding unless the deployment has 50 or fewer users in a rolling 30-day period, the copyright holder grants permission, or an enterprise license is bought. Check that clause with legal before rollout, and review the LICENSE_HISTORY file, since code before a specific commit remains under the earlier BSD-3-Clause terms.
LibreChat
An MIT-licensed, Docker-based chat platform that supports OpenAI, Azure OpenAI, Google, Vertex AI, Anthropic, AWS Bedrock and Ollama, so one interface can route to several model providers. It includes an agent framework with Model Context Protocol support and multi-user authentication through OAuth2, LDAP and email. The trade-off is operational: you own the provider keys, the database, and the decisions on what each provider may receive, so document which model endpoints are approved for which data classes.
AnythingLLM
An MIT-licensed application from Mintplex Labs that runs as a Docker deployment or as a desktop app for macOS, Windows and Linux. It supports multiple users, connects local or cloud LLMs to ingested documents, and includes built-in agents and vector databases. It suits teams that want document chat with a simple setup. Confirm how its user roles, workspace permissions and audit logging meet your access requirements, since these are the areas where small deployments often need extra controls.
Onyx
An open-core platform where the Community Edition is MIT licensed and covers retrieval, AI chat, agents and actions. It indexes more than 50 applications with their source permissions, supports SSO through Google OAuth, OIDC or SAML, and provisions users with SCIM. Code in the ee directories falls under the Onyx Enterprise License, so confirm which features your plan needs sit in the Community Edition. It is the closest match to Glean or Copilot style search, and it is the only option here that documents an air-gappable deployment.
How each ChatGPT Enterprise capability maps to open source
This table maps each ChatGPT Enterprise capability to the open-source option that covers it, with Open WebUI, LibreChat, AnythingLLM and Onyx as the main candidates. Read the cells as a starting point: where coverage is partial, the reader must confirm the feature in the specific release they plan to deploy and build the missing control themselves. The main takeaway is that the audit and export layer is the component most likely to need custom work.
| Capability | Open-source coverage | What to confirm or build | Where it usually falls short |
|---|---|---|---|
| SAML SSO through an identity provider | LibreChat supports OAuth2, LDAP and email authentication. Onyx supports SSO through Google OAuth, OIDC or SAML. Open WebUI and AnythingLLM need checking for your IdP. | Confirm which protocol your identity provider uses and test it against the chosen platform before committing. | Protocol support differs by platform, so SAML is not guaranteed everywhere. Pure LDAP setups may need a broker. |
| Admin retention controls | No direct equivalent is supplied for all four options. Retention is a property of the database and storage you operate. | Define retention periods in your own policy and enforce them with database jobs or storage lifecycle rules. | Default settings may keep conversations indefinitely, so retention must be configured, not assumed. |
| Conversation view and export | Chat history is stored in the platform database. Export and admin-wide viewing depend on the option and version. | Confirm whether admins can view all users' conversations and export them in a usable format. | Admin export may be missing, so teams often write a query-based export against the database. |
| Deletion of conversations | Users can delete their own chats in each platform. Admin-driven deletion across all users and backups must be confirmed. | Verify that deletion removes data from the database, search indexes, vector stores and backups on your schedule. | Vector stores and backups often retain copies after the chat is deleted. |
| Audit logging (Compliance API equivalent) | The Compliance API equivalent is not a supplied feature for these options. Coverage varies by platform and version. | Confirm audit log support per option. Where it is missing, build logging at the reverse proxy, application and database layers. | This is the largest gap. Without it, evidence for access and usage reviews must be assembled manually. |
| Data residency | Residency is controlled by the hosting location rather than a vendor setting. Data stays where you deploy the stack. | Choose the cloud region, data center or air-gapped network, then document where databases, object storage and logs reside. | Model API calls may still leave the region if you point the platform at a cloud model provider. |
| Model choice | LibreChat supports OpenAI, Azure OpenAI, Google, Vertex AI, Anthropic, AWS Bedrock and Ollama. Open WebUI works with Ollama and OpenAI-compatible APIs. Onyx works with Ollama, LiteLLM or vLLM. | Decide whether to use a hosted API, a self-hosted model or both, and record which provider handles each data class. | Switching models can change answer quality and output format, so re-test prompts after each change. |
| No training on your data by default | Self-hosted platforms do not send conversations to a vendor for training. Training behavior depends on the model provider you connect. | Get written terms from each model provider covering training and retention of API inputs and outputs. | A cloud model API used through an open-source front end is still governed by that provider's terms. |
| Document connectors | Onyx indexes 50+ applications with their permissions, including SharePoint, Google Drive, Confluence, Jira and Slack. AnythingLLM and Open WebUI ingest documents with varying connector support. | Confirm which sources you need, check that permissions carry over, and test that a user cannot retrieve documents they cannot access. | Connector coverage and permission sync vary by source and version, so test each one. |
What data control changes when chat runs in your own infrastructure
The controls OpenAI documents are vendor commitments, and you should treat them that way. OpenAI states that it does not train its models on ChatGPT Enterprise business data by default, that deleted conversations are removed from its systems within 30 days unless legally required, and that data residency keeps in-scope content at rest in a selected region. These are useful terms, but they are set by the vendor and governed by your contract. Admins can view, export and delete conversations and reach an audit log through the Enterprise Compliance API, yet the platform itself stays in OpenAI's hosting. Ask your legal and privacy teams which of these commitments appear in your signed agreement, and which are only published on web pages that may change.
In a self-hosted stack such as Open WebUI, LibreChat or AnythingLLM, the chat database, uploaded files, vector stores and logs sit in infrastructure you control. Your own retention schedule decides how long conversations live, how a user deletion propagates to backups, and which logs are kept for investigations. Confirm each of these in your environment: where the chat database is stored, whether backups inherit the deletion policy, and who can read application logs. The control is real only when you can demonstrate it, so plan to test a deletion end to end and keep the evidence.
The model choice decides how much data still leaves your infrastructure. A self-hosted interface does not keep data in-house if every prompt is forwarded to a hosted API provider. Those providers have their own retention and residency terms, which you must review separately. If prompts must stay inside your boundary, route them to a self-hosted model through Ollama or vLLM, or to a model provider that offers an in-region deployment that meets your requirements. Document which model handles each use case, because a single assistant can send different data to different places depending on the model selected.
What the total cost of self-hosted chat includes
This table sets ChatGPT Enterprise contract fees against the components a self-hosted chat stack adds, marking each as one-time or recurring. Read it as a checklist for building a quote, since the main takeaway is that the license line is only one part of the total and the staffing and model lines often decide the result.
| Cost component | Who is paid or who carries it | One-time or recurring | What drives it up or down |
|---|---|---|---|
| ChatGPT Enterprise contract | OpenAI, under the enterprise agreement | Recurring | Seat count, contract term and negotiated terms. Ask for the written order form and what is included. |
| Chat interface software | No license fee for MIT-licensed options like LibreChat and AnythingLLM. Open WebUI has its own license terms to check. | One-time setup, recurring upkeep | Branding rules, user count and whether you need enterprise features. Confirm license terms before rollout. |
| Model inference | API provider per token, or hardware and hosting for self-hosted models | Recurring | Usage volume, model size and context length. Ask for per-token pricing or GPU sizing in writing. |
| Infrastructure hosting | Your cloud account or data center | Recurring | Concurrent users, uptime targets and redundancy. Ask for a sizing estimate tied to your user count. |
| Implementation and integration | Implementation partner or internal team | One-time | Number of identity, logging and data sources connected. Request a fixed-scope deliverable list. |
| Identity integration (SSO and provisioning) | Internal identity team or partner | One-time, with ongoing changes | Use of SAML, OIDC or LDAP and group mapping complexity. Confirm which protocols each option supports. |
| Migration of history and users | Internal team or partner | One-time | Volume of existing conversations and the export path from ChatGPT Enterprise. Confirm what the Compliance API export covers. |
| Backups, upgrades and monitoring | Internal platform team or retainer provider | Recurring | Release cadence, patching windows and uptime needs. Ask who owns on-call and upgrade testing. |
| Internal skills and staffing | Your staff, or external support | Recurring | Team size, prior container or Kubernetes experience and governance workload. Ask for named roles and time commitments. |
What you give up when you leave ChatGPT Enterprise
Moving from a hosted assistant to a self-hosted one changes who runs the platform, who produces compliance evidence and which models you can reach. Each card below names one loss, the workaround that usually covers it and the check to run before you commit.
Vendor-managed compliance reports
ChatGPT Enterprise comes with OpenAI's SOC 2 Type 2 audits, which you can point to during vendor reviews. A self-hosted stack has no such report by default. You build the control evidence yourself (access reviews, logging, backup tests, change records) and have an independent auditor assess it. DeployOpen prepares systems and evidence for that assessment but does not issue audit reports. Ask your auditor which controls they will sample before you design the evidence trail.
Access to OpenAI models and features
New OpenAI features reach ChatGPT Enterprise first, and a self-hosted interface only exposes what its model connectors support. LibreChat supports OpenAI, Azure OpenAI, Anthropic, Google and other providers, and Onyx works with self-hosted models through Ollama, LiteLLM or vLLM. Check that the specific model and feature you rely on are available through the API you plan to use, because a chat interface cannot add capabilities the provider does not expose.
Managed uptime and patching
OpenAI operates ChatGPT Enterprise, including availability, upgrades and incident response. Once you self-host, your team owns the containers, databases, certificates and version upgrades for each chat or search component. Write down who receives alerts, who approves upgrades and how you restore from backup before users depend on the system. Glean's Customer Hosted model shows the same split: the vendor may still manage upgrades and monitoring, which is a different arrangement from running it yourself.
Open WebUI branding clause limits
Open WebUI uses a BSD-3-Clause style license with an added clause that restricts removing or altering its branding. That clause does not apply in every case: it lifts for deployments of 50 or fewer users in a rolling 30-day period, when the copyright holder grants permission, or when an enterprise license is bought. Code before a specific commit remains under the earlier BSD-3-Clause terms. Larger teams should confirm with the license file and their counsel whether their rollout needs an enterprise license or a branding exception.
Enterprise-only Onyx directory features
Onyx Community Edition is MIT licensed and covers retrieval, chat, agents and actions, but code in the ee directories falls under the Onyx Enterprise License. Connector and permission features that live there may not be available under the community terms. Before you plan a search rollout across SharePoint, Confluence or Slack, map each permission sync and SSO requirement to the edition that includes it, and confirm the licensing with Onyx.
Polished admin tooling that varies
ChatGPT Enterprise gives workspace admins a managed console for retention settings, user conversation export and deletion. Open-source options differ widely here. AnythingLLM and Open WebUI offer multi-user administration, while retention, audit trails and export workflows may need scripts, database queries or added logging. List the admin tasks your compliance team performs each month and test each option against them in a pilot before you choose.
Which option fits by situation
When the main requirement is that conversations and uploaded files stay in a region or account you choose, a self-hosted interface running in your own cloud subscription or data center is the most direct route. ChatGPT Enterprise offers regional data residency for stored content, but that control applies inside OpenAI's hosted service. A self-hosted stack moves the storage location under your own controls.
- Storage location: Databases, uploads and vector indexes sit in the account or region you provision, so residency is set by your architecture.
- Model routing: Pair the interface with a model endpoint in the same region, or with a self-hosted model, so prompts do not leave the boundary you define.
- Logging: Application logs and chat histories are stored in your systems, so retention and deletion follow your own policy.
- Operations: Your team or a partner patches, backs up and monitors the stack, which is work ChatGPT Enterprise removes from you.
Before committing, confirm with each model provider whether its API keeps prompts in the region you require and what it logs. Check that your residency rule covers embeddings, backups and telemetry, not only the primary database.
Regulated organizations and networks without internet egress usually need the whole chat path, including the interface, retrieval and model, to run inside a controlled boundary. Open WebUI can run entirely offline when paired with Ollama, which makes it a common starting point for isolated environments. LibreChat suits teams that need multi-user authentication through OAuth2, LDAP or email in a Docker deployment.
- Offline operation: Open WebUI works with local models through Ollama, so no prompt needs to leave the network.
- Directory integration: LibreChat supports LDAP and OAuth2, which matches many internal identity setups.
- Retrieval in isolation: Onyx documents an air-gappable deployment with its document index and processing inside a self-contained set of services.
- Evidence: You must produce access logs, change records and configuration baselines yourself for any assessment.
Before committing, confirm that every container image, model file and package can be mirrored into your environment, and ask your compliance advisor which controls your assessor expects you to evidence. DeployOpen prepares systems and evidence for assessment, but independent auditors issue any report.
Cost-driven teams usually compare a per-seat subscription with a model that combines infrastructure, model usage and operations. The open-source interfaces themselves carry no license fee in most cases, so the real comparison is the recurring cost of hosting, model tokens or model hardware, and the staff time needed to run the platform. Open WebUI and LibreChat are free to deploy, and AnythingLLM is MIT licensed, so your spend moves to infrastructure and support.
- One-time costs: Setup, identity integration, connector configuration and initial testing.
- Recurring costs: Compute, storage, model API usage or GPU capacity, backups and staff time for upgrades.
- Cost drivers: Number of active users, model size, token volume and how long chat history is retained.
- Quote requests: Ask for fixed scope, named deliverables and which items are excluded from the price.
Before committing, model your expected usage against the hosted seat price in your current contract and against the infrastructure you already run. Check whether Open WebUI's branding clause applies to your user count, because it affects whether you may change the interface branding at scale.
Teams that want to switch between model vendors, or build custom agents and tool integrations, benefit from interfaces that treat providers as interchangeable. LibreChat supports OpenAI, Azure OpenAI, Google, Vertex AI, Anthropic, AWS Bedrock and Ollama, and includes an agent framework with Model Context Protocol support. AnythingLLM connects local or cloud LLMs and ships with built-in agents and vector database options.
- Provider choice: Route different teams to different models without rebuilding the interface.
- Agent tooling: LibreChat's Model Context Protocol support lets you connect internal tools through a common interface.
- Vector store choice: AnythingLLM lets you select the vector database that fits your operations skills.
- Code ownership: Open-source code lets you modify behavior, though you take on the maintenance of those changes.
Before committing, confirm which providers your legal team has approved and how each one handles data retention. Test each agent or tool integration against your permission model, because a custom tool can expose data the interface itself would not show.
When the main job is answering questions from company content such as SharePoint, Confluence, Google Drive or Slack, a chat window alone is not enough. Onyx is the strongest fit here because it combines retrieval, chat and connectors. Its Community Edition is MIT licensed and covers RAG, AI chat and agents, and the repository includes connectors for sources such as SharePoint, OneDrive, Confluence, Jira and Slack.
- Connector coverage: Check that each source you need has a connector in the current repository before you plan around it.
- Permission handling: Onyx indexes applications with their permissions and supports SSO through Google OAuth, OIDC or SAML and SCIM provisioning. Test that answers respect access rights in each source.
- Enterprise features: Code in the ee directories falls under the Onyx Enterprise License, so confirm which features your deployment needs.
- Model choice: Onyx works with self-hosted models through Ollama, LiteLLM or vLLM.
Before committing, run a permission test with two users who have different access, and confirm that search results match what each user can open in the source system.
How to move off ChatGPT Enterprise without losing conversations
This sequence moves conversation history, identities and workflows from ChatGPT Enterprise to a self-hosted assistant in your own infrastructure while the subscription still gives you export access. It ends with a verified cutover and documented deletion of the OpenAI-side data.
Inventory users, workspaces, GPTs and integrations
List every workspace, member, custom GPT, connected tool and API integration that touches ChatGPT Enterprise, and record who owns each one. Identify which teams depend on specific GPTs or shared projects, because these need a replacement design rather than a simple export. The step is complete when the owner of each item is named and each integration has a target in your new stack or a retirement decision.
Export conversations through the Compliance API and admin tools
While the subscription is active, use the Enterprise Compliance API and workspace admin export to pull conversation records and the audit log of conversations and GPTs. Confirm with OpenAI which export formats, date ranges and user scopes your plan supports, and whether any data is excluded. Store exports in your own encrypted repository with access logging. The step is complete when you can open a sample of exported conversations for each team and reconcile counts against the admin console.
Map identity and roles to the new platform
Document how users sign in today through SAML SSO and which admin, member and viewer roles exist. Then define equivalent roles in the self-hosted platform, choosing an identity path such as OIDC, SAML, LDAP or OAuth2 that your directory supports. LibreChat lists OAuth2, LDAP and email authentication, and Onyx lists SSO through Google OAuth, OIDC or SAML with SCIM provisioning, so confirm the option that matches your identity provider. The step is complete when a test account in each role signs in and sees only the expected workspaces.
Pilot with a small group
Choose a limited group of users with representative tasks, such as drafting, document summarization and coding help, and move them to the new assistant. Connect the model providers or self-hosted models you have approved, and record which tasks need prompt or model changes. Collect structured feedback on answer quality, latency and missing features. The step is complete when the pilot group has completed its agreed task list and the open issues are logged with owners.
Run in parallel and compare answers
Keep ChatGPT Enterprise available while the pilot group and reference prompts run in both systems. Compare answers for accuracy, citations, refusals and formatting against a fixed set of prompts that reflect real workloads. Record differences and decide which are acceptable, which need configuration, and which block cutover. Because ChatGPT Enterprise is OpenAI-hosted only, this comparison is the main evidence for whether the new platform meets your needs. The step is complete when the parallel results are signed off by the business owners.
Cut over users and retire the old workflows
Announce a cutover date, move all remaining users and shared assets to the new platform, and update any integrations that called the ChatGPT Enterprise environment. Keep the exported archive available for the period your records policy requires. Remove shared links, custom GPT references and API keys that pointed to the old workspace. The step is complete when no active production workflow depends on ChatGPT Enterprise and the support queue has no unresolved migration tickets.
Decommission and confirm deletion
Confirm with OpenAI, in writing, how it removes your Enterprise data after you end the subscription and what it will certify about deletion. OpenAI's privacy page describes deleted conversations being removed within 30 days unless legally required, and admin retention settings control this. Check your contract for renewal notice periods and data return terms before the renewal date. The step is complete when you hold OpenAI's written deletion confirmation and your internal records show the archive location and retention owner.
Questions teams ask before replacing ChatGPT Enterprise
How much does self-hosting cost compared with a ChatGPT Enterprise subscription?
The answer depends on components, not on a single line item. A ChatGPT Enterprise subscription bundles the software, hosting, model access and vendor operations into one per-seat contract, so the comparison has to be built from the parts you would pay for yourself. Self-hosting costs include infrastructure in your cloud account or data center, GPU or API spend for models, storage and backups, identity integration, and staff time for upgrades, monitoring and security patching. Model usage is often the largest recurring driver and it scales with how heavily people use the tool. Ask any implementation partner for a quote that separates one-time work (planning, integration, hardening, documentation) from recurring items (hosting, model consumption, support retainers), and state the usage assumptions behind each number so you can rerun the model with your own headcount and workload.
How long does a migration from ChatGPT Enterprise take?
Timing is driven by scope, not by the chat interface itself, which can be installed quickly. The main variables are how many identity providers and user groups you connect, whether you need conversation history or only new chats, how many models you must validate, and whether you also need document search over company content. Compliance evidence, such as access reviews and retention settings, adds time if your assessors expect it to be documented before cutover. A practical approach is a short pilot with a small user group, followed by staged rollout. Your provider should agree timelines and deliverables in writing before work starts, and you should confirm how each phase is scoped and what your own team must supply.
Can our own engineering team run a self-hosted AI chat platform?
Many teams can run the core platform with existing container and identity skills, but the work is ongoing rather than a single installation. Open WebUI installs with Docker, Python pip or Kubernetes, and LibreChat is self-hosted with Docker, so a platform team familiar with those tools has a clear starting point. The harder parts are operating the model layer, keeping components patched, monitoring usage, managing backups, and responding to security findings. Ask yourself who will own upgrades, on-call coverage and model provider credentials. If those responsibilities are not clear, a support retainer can cover upgrades and monitoring while your team keeps control of the environment.
Does open-source software keep our data inside our own infrastructure?
Open-source software keeps data where you deploy it, but only if the whole chain is under your control. Chat history, uploaded files, vector indexes and logs stay in your environment when the platform runs in your cloud account, data center or air-gapped network. Data still leaves your environment if you call a hosted model API, so the data path depends on which model you choose. Onyx documents an air-gappable deployment that works with self-hosted models through Ollama, LiteLLM or vLLM, and Open WebUI can run entirely offline with Ollama. Confirm the outbound network rules, the model endpoints in use, and where backups and telemetry are sent during your design review.
Can we keep SSO and SCIM provisioning after leaving ChatGPT Enterprise?
Yes, the open-source options differ in what they support, so check each against your identity provider. ChatGPT Enterprise documents SAML SSO, and LibreChat offers multi-user authentication with OAuth2, LDAP and email. Onyx supports SSO through Google OAuth, OIDC or SAML and provisions users with SCIM. Open WebUI and AnythingLLM need the same review before you commit. Confirm which protocols your identity provider uses, how group membership maps to roles, how deprovisioning works when someone leaves, and whether automated provisioning is available for the platform you choose. Test those flows in a pilot before rollout.
What happens to exported conversations and the Compliance API data?
ChatGPT Enterprise admins can export conversations and access an audit log through the Enterprise Compliance API, which is the documented export path for that product. After migration, that export becomes an input to your own retention and archive process rather than a vendor feature. Decide which records you must keep, for how long, and in what format, then confirm the platform can store chat history, user and admin actions, and deletion events in a way your records team accepts. Plan for conversations that contain personal or regulated data, and document how deletion requests are handled across backups and vector indexes.
How does the Open WebUI license apply to a larger rollout?
Open WebUI is licensed under the Open WebUI License, a BSD-3-Clause style license with an added branding clause. The clause forbids removing or altering Open WebUI branding unless the deployment has 50 or fewer users in a rolling 30-day period, the copyright holder grants permission, or an enterprise license is bought. A larger rollout therefore needs one of those conditions met if you intend to change branding. Code before a specific commit remains under the earlier BSD-3-Clause terms, so check the LICENSE_HISTORY file for the version you deploy. Confirm the exact terms with the copyright holder or legal counsel before you scale, because a license reading is not legal advice.
Will DeployOpen certify that our deployment is compliant?
No. DeployOpen plans, implements, hardens, documents and supports self-hosted platforms and prepares the systems, controls and evidence that compliance assessments examine. It does not issue audit reports, attestations or certifications. Independent auditors and certification bodies perform those assessments. Engagements are fixed-scope, with scope, deliverables and timelines agreed in writing before work starts, and a quote follows from that written scope. Ask any provider to name which controls and evidence they will prepare, and confirm with your auditor which items they need.
How DeployOpen scopes, migrates and supports the replacement
DeployOpen plans, implements, integrates, hardens, backs up, documents and supports self-hosted open-source chat platforms inside your own cloud accounts, data centers or air-gapped networks. For ChatGPT Enterprise replacements, the work starts with a written scope covering identity, retention, export and the model providers you choose, so that the team knows exactly what is delivered before any build begins.
How we scope
- Written scope agreed before work starts, covering users, identity provider, model providers and hosting location.
- Mapping of current ChatGPT Enterprise usage, retention settings and conversation export needs to target controls.
- Decisions on which interface fits your needs, such as Open WebUI, LibreChat or AnythingLLM, documented in the scope.
How we build
- Deployment in your infrastructure using Docker, Python or Kubernetes, as the chosen platform supports.
- Configuration of SSO, role mapping and retention settings, with testing of sign-in and access behavior.
- Integration with self-hosted models or the API providers you approve, plus security testing of the deployed stack.
How we hand over
- Documentation of architecture, configuration, backup procedures and the controls prepared for compliance review.
- Knowledge transfer to your platform and security teams so they can operate and audit the system.
- Evidence packages prepared for your assessors, noting that independent auditors and certification bodies issue any reports.
How we support
- Ongoing support, upgrades and continuous testing available as a retainer after handover.
- Patching and version upgrades of the chat platform, model connectors and supporting services on agreed terms.
- Confirm with us in the quote which license terms, such as Open WebUI branding rules, apply to your user count.
