> ## Documentation Index > Fetch the complete documentation index at: https://docs.streamkap.com/llms.txt > Use this file to discover all available pages before exploring further. # Suggest Agent System Prompt > Generate a starter ``customInstructions`` block for the wizard. Same two credential shapes as ``/agents/validate-llm``: - **Inline** - raw ``apiKey`` (+ optional ``baseUrl``) supplied directly. - **Saved connection** - ``savedConnectionId`` references a stored ``llmConnection``; BE loads decrypted ``apiKey`` / ``baseUrl`` / ``provider`` server-side so the browser never holds the plaintext. Composes a meta-system-prompt + a structured user message describing the agent (type, name, description, topics, fields, tools, output schema, optional existing draft) and calls the user's own LLM. The runtime ``build_system_prompt`` already emits the per-agentType base template + schema field listing + tool signatures at deploy time, so the LLM is constrained to produce only the user-facing instructions slot - the meta-prompt enforces that explicitly. Nothing is persisted. Per-tenant rate limit: dedicated ``BUCKET_PROMPT_SUGGEST`` bucket (30/min/tenant) separate from ``validate-llm``, so heavy "Generate" usage (humans iterating on prompt drafts) doesn't eat into the credential-check budget. Provider failures (timeouts, 4xx, schema errors) surface as 400 with the provider's error message so the FE's existing error-extraction works unchanged. ## OpenAPI ````yaml /openapi/openapi.json post /agents/system-prompt-suggestion openapi: 3.1.0 info: title: Streamkap REST API description: >- The Streamkap REST API allows you to programmatically manage your CDC pipelines, sources, destinations, transforms, and more. Authenticate using a bearer token obtained from the [Access Token](/api-reference/auth/access-token) endpoint with your API client credentials. contact: name: Streamkap Support url: https://streamkap.com/ email: support@streamkap.com license: name: Proprietary version: 2.0.0 servers: - url: https://api.streamkap.com description: Production security: [] tags: - name: Agents description: Create, deploy, and manage Flink-based AI agents. - name: Agents Observability description: >- Read-only views over the MCP audit log: agent rollup, sessions, spans, histograms, external-agent verify. - name: Alerts description: Manage alert subscribers, preferences, and notification credentials. - name: Authentication description: Obtain and refresh access tokens, manage client credentials and roles. - name: Billing description: Retrieve usage metrics, summaries, and export billing data. - name: Consumer Groups description: List, inspect, and reset Kafka consumer group offsets. - name: Dashboard description: Retrieve organisation-level statistics and overview data. - name: Destinations description: Create, configure, and manage data destinations and their lifecycle. - name: Kafka Access description: Manage Kafka users and their access permissions. - name: Logs description: Query and summarise system logs. - name: Pipelines description: Create, configure, and manage CDC pipelines and their lifecycle. - name: Project Keys description: Manage project keys that bundle API credentials and optional Kafka access. - name: Schema Registry description: Browse schema subjects, versions, and retrieve schema definitions. - name: Services description: View and switch service metadata. - name: Sources description: >- Create, configure, and manage data sources, snapshots, and their lifecycle. - name: Tags description: Create, update, and manage resource tags for organisation. - name: Topics description: Browse topic details, statistics, configurations, metrics, and messages. - name: Transforms description: >- Create, deploy, and manage data transforms, unit tests, and implementation details. paths: /agents/system-prompt-suggestion: post: tags: - Agents summary: Suggest Agent System Prompt description: >- Generate a starter ``customInstructions`` block for the wizard. Same two credential shapes as ``/agents/validate-llm``: - **Inline** - raw ``apiKey`` (+ optional ``baseUrl``) supplied directly. - **Saved connection** - ``savedConnectionId`` references a stored ``llmConnection``; BE loads decrypted ``apiKey`` / ``baseUrl`` / ``provider`` server-side so the browser never holds the plaintext. Composes a meta-system-prompt + a structured user message describing the agent (type, name, description, topics, fields, tools, output schema, optional existing draft) and calls the user's own LLM. The runtime ``build_system_prompt`` already emits the per-agentType base template + schema field listing + tool signatures at deploy time, so the LLM is constrained to produce only the user-facing instructions slot - the meta-prompt enforces that explicitly. Nothing is persisted. Per-tenant rate limit: dedicated ``BUCKET_PROMPT_SUGGEST`` bucket (30/min/tenant) separate from ``validate-llm``, so heavy "Generate" usage (humans iterating on prompt drafts) doesn't eat into the credential-check budget. Provider failures (timeouts, 4xx, schema errors) surface as 400 with the provider's error message so the FE's existing error-extraction works unchanged. operationId: suggestAgentSystemPrompt requestBody: content: application/json: schema: $ref: '#/components/schemas/SystemPromptSuggestionRequest' required: true responses: '200': description: Successful Response content: application/json: schema: $ref: '#/components/schemas/SystemPromptSuggestionResponse' '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' security: - FronteggHTTPAuthentication: [] components: schemas: SystemPromptSuggestionRequest: properties: provider: $ref: '#/components/schemas/AgentLlmProviderEnum' model: type: string maxLength: 200 title: Model default: '' apiKey: type: string maxLength: 512 title: Apikey description: Raw key - never stored default: '' baseUrl: anyOf: - type: string maxLength: 512 - type: 'null' title: Baseurl savedConnectionId: anyOf: - type: string maxLength: 100 - type: 'null' title: Savedconnectionid description: >- If set, BE resolves provider/apiKey/baseUrl from the tenant's saved llmConnection by id and ignores the inline ``apiKey`` / ``baseUrl``. agentType: type: string maxLength: 50 minLength: 1 title: Agenttype description: Agent type (e.g. react, workflow) agentName: anyOf: - type: string maxLength: 100 - type: 'null' title: Agentname description: anyOf: - type: string maxLength: 500 - type: 'null' title: Description inputTopics: items: type: string type: array maxItems: 20 title: Inputtopics selectedFields: items: type: string type: array maxItems: 50 title: Selectedfields tools: items: $ref: '#/components/schemas/ToolContext' type: array maxItems: 50 title: Tools description: >- Tool name + description list. Replaces the older toolNames-only shape. outputSchema: items: $ref: '#/components/schemas/OutputSchemaField' type: array maxItems: 50 title: Outputschema filterSQL: anyOf: - type: string maxLength: 4000 - type: 'null' title: Filtersql description: >- Active record-filter (matches AgentInputConfig.filterSQL). Anchors the LLM's prompt to the actual shape of records reaching the agent. existingCustomInstructions: anyOf: - type: string maxLength: 8000 - type: 'null' title: Existingcustominstructions description: Existing draft to refine. Matches the FE textarea cap. generationHint: anyOf: - type: string maxLength: 500 - type: 'null' title: Generationhint description: >- Optional one-line steer the user types into the wizard's 'Hint for AI' input. The meta-prompt already gets the agent's metadata, so this should carry use-case intent (tone, edge-case priority, what to emphasize) rather than restate the data shape. additionalProperties: false type: object required: - provider - agentType title: SystemPromptSuggestionRequest description: >- Request body for ``POST /agents/system-prompt-suggestion``. Mirrors ``ValidateLlmRequest`` for the credential half (inline apiKey or savedConnectionId, with the same XOR rule) and adds the agent-context fields the meta-prompt needs to draft a starter ``customInstructions`` block. The runtime ``build_system_prompt`` already injects the per-agentType base template, schema field listing, and tool signatures - so the LLM is asked to produce *only* the user-facing instructions slot, never the boilerplate wrapper. See ``app.services.agent_prompt_suggester`` for the meta-prompt that enforces that. SystemPromptSuggestionResponse: properties: generatedPrompt: type: string title: Generatedprompt tokensIn: type: integer title: Tokensin tokensOut: type: integer title: Tokensout model: type: string title: Model provider: type: string title: Provider additionalProperties: false type: object required: - generatedPrompt - tokensIn - tokensOut - model - provider title: SystemPromptSuggestionResponse description: |- Response body for ``POST /agents/system-prompt-suggestion``. Token counts come straight from the provider's ``usage`` block (zeroed when the provider didn't return one). ``model`` / ``provider`` echo back the resolved values so the FE can display "generated with anthropic / claude-sonnet-4-5" alongside the suggestion. HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' type: array title: Detail type: object title: HTTPValidationError AgentLlmProviderEnum: type: string enum: - anthropic - openai - openai-responses - ollama - azure - azure-openai - bedrock - qwen - openai-compatible title: AgentLlmProviderEnum description: >- Unified LLM provider enum. A single ``AgentLlmConnection`` row carries one provider and a set of capabilities (chat / embedding). ``PROVIDER_CAPABILITIES`` below pins which capabilities each provider can serve — picked by the FE Connections drawer and re-validated server-side on every write. ToolContext: properties: name: type: string maxLength: 200 minLength: 1 title: Name description: Tool name as referenced in the agent description: type: string maxLength: 1000 title: Description description: What the tool does, free-form default: '' additionalProperties: false type: object required: - name title: ToolContext description: >- One tool entry in the suggestion payload. Carries name + description so the LLM can reason about *when to call which tool*. Earlier shape passed names only, which gave the LLM no grounding for tool-use advice ("call lookup_user when..." with no idea what lookup_user actually does). Description is optional - tools without one render as "(no description)" so the meta-prompt still sees the name. OutputSchemaField: properties: name: type: string maxLength: 100 minLength: 1 title: Name description: Output field name type: type: string maxLength: 50 minLength: 1 title: Type description: Output field type as the wizard renders it additionalProperties: false type: object required: - name - type title: OutputSchemaField description: >- One row of the output schema as the wizard's prompt-suggestion form sees it. Loosely typed - the suggestion endpoint only needs name/type prose for the meta-prompt, so we don't reuse ``AgentOutputSchemaTypeEnum`` here. That keeps the suggestion path tolerant of future schema-type additions without forcing a model bump on the read-only suggestion surface. ValidationError: properties: loc: items: anyOf: - type: string - type: integer type: array title: Location msg: type: string title: Message type: type: string title: Error Type input: title: Input ctx: type: object title: Context type: object required: - loc - msg - type title: ValidationError securitySchemes: FronteggHTTPAuthentication: type: http scheme: bearer ````