> ## Documentation Index > Fetch the complete documentation index at: https://docs.sreagent.app/llms.txt > Use this file to discover all available pages before exploring further. # Create an AI provider > Add an AI provider this organization's own key pays for. provider is one of anthropic, openai, openrouter, ollama, bedrock, gemini, portkey, langchain. model_overrides pins a model per purpose, keyed by purpose: investigation, suggestion, report_generation, alert_grouping, context_compaction, pr_review, remediation, oversight, config_watch, finops, slack_assistant, board_assistant, runbook_suggestion, sli_suggestion, capacity_planning, status_page_draft, ticket_synthesis, security_triage. Needs the plan's bring-your-own-provider feature. ## OpenAPI ````yaml /api-reference/openapi.json post /ai_providers openapi: 3.1.0 info: description: >- A resource-shaped endpoint onto the same tool surface the MCP server (/api/mcp) reaches. Every gate, refusal and audit row a caller sees here is the exact one the MCP endpoint answers for the same tool. Authenticated with an API key holding the api:admin scope, or the read-only api:config_read scope for GET alone. title: SRE Agent configuration API version: 1.0.0 servers: - url: https://sreagent.app/api/v1/config security: - apiKey: [] tags: - name: ai_providers - name: ai_settings - name: alert_mutes - name: alert_routes - name: aws_external_id - name: certificate_monitors - name: change_notifications - name: compliance_periods - name: connectors - name: data_sources - name: deploy_policies - name: export - name: github_settings - name: image_targets - name: notification_settings - name: organization_settings - name: outbound_configs - name: outbound_rules - name: overseer_settings - name: prompt_templates - name: repo_settings - name: service_bindings - name: slack - name: slis - name: slos - name: status_page_components - name: status_page_incidents - name: status_page_settings - name: synthetic_checks - name: team_members - name: teams - name: ticket_import_rules - name: ticket_integrations paths: /ai_providers: post: tags: - ai_providers summary: Create an AI provider description: >- Add an AI provider this organization's own key pays for. provider is one of anthropic, openai, openrouter, ollama, bedrock, gemini, portkey, langchain. model_overrides pins a model per purpose, keyed by purpose: investigation, suggestion, report_generation, alert_grouping, context_compaction, pr_review, remediation, oversight, config_watch, finops, slack_assistant, board_assistant, runbook_suggestion, sli_suggestion, capacity_planning, status_page_draft, ticket_synthesis, security_triage. Needs the plan's bring-your-own-provider feature. operationId: create_ai_provider parameters: [] requestBody: content: application/json: schema: properties: api_key: description: >- The provider's key. Write-only: it is never answered by any tool here, an omitted one keeps the stored key, and an empty one is refused rather than read as a way to remove it. To remove it, delete the provider. Paste the key itself; a URL goes in base_url and is refused here. format: password type: string writeOnly: true auto_upgrade_model: description: >- Whether the platform may move `model` on to a successor once the configured one drops out of the provider's list. Off by default: a pinned model stays pinned. type: boolean base_url: description: >- Where to send the calls, for a gateway or a self-hosted endpoint. null uses the provider's own. type: - string - 'null' enabled: description: >- Whether the router may route to this provider at all. A disabled row keeps its key and its pins and is skipped. type: boolean max_tokens: description: >- The ceiling on one answer's length. null uses the platform's own. type: - integer - 'null' model: description: >- The model to use, spelled as the provider spells it. null leaves the choice to the platform's own default. type: - string - 'null' model_overrides: description: >- A model per purpose, keyed by purpose: investigation, suggestion, report_generation, alert_grouping, context_compaction, pr_review, remediation, oversight, config_watch, finops, slack_assistant, board_assistant, runbook_suggestion, sli_suggestion, capacity_planning, status_page_draft, ticket_synthesis, security_triage. Merged into what is stored, so a map naming one purpose leaves the rest alone; an empty string under a purpose clears that pin. The value "auto" hands the choice to the platform's own model selection for that purpose. type: object name: description: >- What this provider is called here, unique within the organization. It is a label for the people reading the list, not anything the provider sees. type: string priority: description: >- The order the router tries this organization's providers in, lowest first. At least 1. type: integer provider: description: >- Which provider this row talks to: anthropic, openai, openrouter, ollama, bedrock, gemini, portkey, langchain. enum: - anthropic - openai - openrouter - ollama - bedrock - gemini - portkey - langchain type: string rate_limit_rpm: description: >- Requests per minute this provider will accept. null means no limit. type: - integer - 'null' temperature: description: 0 to 2. null leaves the provider's own default in place. type: - number - 'null' timeout_ms: description: >- How long one request may take, at most 600000. null uses the platform's own. type: - integer - 'null' required: - name - provider type: object required: true responses: '201': description: Created. '409': description: A row already matches this resource's natural key. '422': description: The tool refused the request's shape or content. default: description: The request failed. content: application/json: schema: $ref: '#/components/schemas/Error' components: schemas: Error: type: object description: The body of every failed request. properties: error: type: string description: >- A short machine-readable code such as forbidden, conflict or read_only_key. message: type: string description: A sentence that says what to change. required: - error - message additionalProperties: true securitySchemes: apiKey: description: >- An sre_ak_* API key holding the api:admin scope (or api:config_read, which every write refuses with read_only_key). scheme: bearer type: http ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.