Skip to content

Enterprise > Enterprise features

Team-managed LLM API keys and endpoints

Configure shared LLM provider API keys and custom endpoints for your team from the Admin Panel, including cloud agents.

Open in ChatGPT ↗
Ask ChatGPT about this page
Open in Claude ↗
Ask Claude about this page
Copied!

Team admins can configure shared provider API keys and OpenAI-compatible endpoints for local and cloud Warp Agent runs. Team members select the configured models without handling the credentials themselves.

Unlike personal BYOK and personal endpoints, team-managed configuration is stored securely by Warp and is available to cloud agents.

ProductPlans that include team-managed keys and endpoints
Warp Terminal and local agentsBusiness and Enterprise
Warp FactoriesScale and Enterprise

For Warp Factories, these credentials configure inference for the Warp Agent, not a third-party harness. See harness authentication for those credentials.

  • Admin-configured, shared across the team - An admin sets provider keys and custom endpoints once in the Admin Panel, and they appear automatically in every member’s model picker.
  • Local and cloud runs - Use your team’s providers with the Warp Agent in either location.
  • First-party keys and custom endpoints - Configure API keys for OpenAI, Anthropic, and Google, and add one or more OpenAI-compatible custom endpoints (for example, OpenRouter or LiteLLM).
  • Optional personal configuration - Admins choose whether members may also add their own API keys and endpoints. See member key precedence for how Warp chooses a key.

How team-managed providers differ from self-serve BYOK and BYOLLM

Section titled “How team-managed providers differ from self-serve BYOK and BYOLLM”
NameConfigured byStoredWorks with cloud agentsPlans
Team-managed API keys and endpointsTeam admin (Admin Panel)Server-side by WarpYesSee eligibility by product
Personal BYOKEach memberLocally on the member’s deviceNoFree and eligible paid plans for Warp Terminal and local agents
Personal custom inference endpointEach memberLocally on the member’s deviceNoFree and eligible paid plans for Warp Terminal and local agents
BYOLLM — AWS BedrockTeam admin (Admin Panel)Cloud-native IAM (no stored keys)YesEnterprise agreement
BYOLLM — Gemini EnterpriseTeam admin (Admin Panel)Cloud-native IAM (no stored keys)No (interactive sessions only)Enterprise agreement

Choose team-managed keys and endpoints to route through provider APIs (OpenAI, Anthropic, Google) or an OpenAI-compatible endpoint using stored keys. Under an Enterprise agreement, AWS Bedrock BYOLLM instead uses cloud-native IAM routing through your AWS account, including cloud agent support. Gemini Enterprise BYOLLM routes through Vertex AI in your Google Cloud project for interactive sessions only.

  • First-party keys (BYOK) - The admin adds an API key for a provider Warp supports (OpenAI, Anthropic, or Google). Members select the normal Warp model for that provider. Routing follows member key precedence, including the Claude linked-billing exception.
  • Custom endpoints (BYOE) - The admin adds one or more OpenAI-compatible Chat Completions endpoints. Each endpoint has a name, a base URL, an API key, and one or more models, where each model has a model name (sent to the endpoint) and an optional alias (shown to members). Team endpoint models appear in every member’s model picker as their own entries, labeled with the endpoint name.

For local runs, a permitted personal provider key normally takes precedence over the team’s key. If the team enables Claude linked billing, members can’t override it with personal Anthropic keys. To return to key-based routing, an admin must stop linked billing for the team.

  • First-party keys - Outside team-enabled Claude linked billing, a permitted personal key is used first, then the team key, then Warp-managed inference if allowed.
  • Custom endpoints - A member’s own endpoint and a team endpoint never collide, even if they share a name. The member’s endpoint is used when they select it; team endpoints resolve to the team’s stored credential.

Cloud Warp Agent runs can use your team’s keys and endpoints when you select a configured provider model or endpoint. Auto uses Warp-provided inference; see billing behavior. Personal keys and endpoints stored on a member’s device aren’t copied to cloud runs.

Configuring team-managed keys and endpoints

Section titled “Configuring team-managed keys and endpoints”

Configure team providers from the Admin Panel Models page, where a First-party API keys card and a Custom endpoints card sit alongside the Direct API and AWS Bedrock cards.

  1. In the Admin Panel, go to the Models page.
  2. To add a team API key, use the First-party API keys card: enter the key for a provider (OpenAI, Anthropic, or Google) and save it. Each provider row shows whether a key is configured without revealing the stored value.
  3. To add a team custom endpoint, use the Custom endpoints card: enter a name, a base URL, an API key, and at least one model (a model name plus an optional alias), then save. An endpoint can’t be saved without a name, a valid base URL, an API key, and at least one model.
  4. Choose whether members may add their own keys and endpoints. When this is off, Warp won’t use any key or endpoint a member adds, even if they’ve saved one — their saved keys are preserved for when it’s turned back on.

The team’s enabled models appear in members’ model pickers when their team settings refresh.

Configuring team providers doesn’t remove Warp-managed models from the model picker. If your plan permits changing Direct API access, turn it off on the Admin Panel’s Models page to remove those models from the picker.

This doesn’t disable member-managed keys and endpoints. Review that separate setting in Configure keys and endpoints if members must use team-managed credentials.

  • Custom endpoints - Each enabled team endpoint model appears in the model picker as its own entry, showing the model’s name (or alias) and the endpoint name. Members select it like any other model.
  • First-party keys - Members select the standard model for a provider (for example, a Claude, GPT, or Gemini model). Routing follows member key precedence and the Claude linked-billing exception.
  • Provider-billed inference - Your provider or endpoint bills team-routed inference directly rather than consuming Warp-provided inference usage.
  • Claude linked billing - When enabled for the team, eligible Claude requests use the linked Claude organization’s billing instead of personal or team Anthropic keys. See linked-billing fallback and charges.
  • Auto uses Warp-provided inference - Auto consumes Warp usage. Select a specific model or endpoint to use your team’s provider. Custom routers apply provider keys after choosing a model; local runs follow member key precedence. Routers cannot target endpoint models.
  • Other Warp charges remain - Customer-supplied inference incurs platform charges according to the product and plan billed for the run. Cloud runs on Warp-hosted compute also incur compute charges.

External inference bills are not fully represented in Warp’s totals. See usage and billing for cost-reporting limitations.

Warp encrypts team API keys and endpoint URLs at rest. Members receive model names, endpoint names, aliases, and whether a provider key is configured, not the stored keys or endpoint URLs.

Warp uses the stored credentials to call your provider without delivering them to member devices or cloud agent environments.

Requests sent through your team’s keys or endpoints are subject to your providers’ terms, not Warp’s ZDR agreements. You’ll need to verify retention with your provider for your specific account, model, and configuration. If your endpoint uses other model providers, verify their retention policies too. Warp cannot enforce provider ZDR for these requests.

Team admins manage content collection for service improvement through “UGC Data Collection” in the Admin Panel’s Privacy settings. Disabling it doesn’t stop storage of cloud agent transcripts, attachments, or artifacts.

Review run-data storage options separately. For factory runs, follow factory ZDR guidance.