Cookbook

Production recipes for the Gateway API

Copy, paste, and ship. Recipes cover prompting systems, agent workflows, multimodal pipelines, and eval guardrails.

Chat completions

Structured prompts with deterministic output and usage totals.

Streaming chat

SSE streaming with chunk-by-chunk updates.

Audio transcription

Upload WAV or stream PCM over WebSocket.

Text to speech

Generate audio with voice, emotion, and speed controls.

Avatar streaming

Synchronized audio + blendshape frames over WebSocket.

Fine-tuning jobs

Create and manage model fine-tuning pipelines.

Prompting systems

System prompts, output contracts, and prompt tests.

Agent workflows

Tool routing, handoffs, and memory patterns.

Guardrails + evals

Policy checks, regression suites, and eval flywheels.

Prompting system starter kit

Build system + developer prompts with contracts, tool gating, and JSON outputs.

You are the Disruptive Rain Ops Analyst.
Mission: deliver an executive brief under 200 words.
Rules:
- Use plain language with no internal policy references.
- If data is missing, say "unknown" and propose the next best action.
Return JSON with keys: summary, risks, next_steps.

System promptsOutput schemasPrompt tests

Prompting checklist

Pin the role, audience, and mission in the system prompt.
Declare non-negotiables such as safety, compliance, or confidentiality.
Provide an output schema and example to eliminate ambiguity.
Add tool usage rules and required data sources.
Include a verification step before final output.

Prompt patterns

Prompt contract

Define output fields, ordering, and token limits.

Few-shot anchor

Include one high-quality example for format alignment.

Tool gating

Specify when to call search, retrieval, or tools.

Critique pass

Validate the draft against requirements before finalizing.

Memory scoping

Reference only approved context windows and metadata.

Fallback plan

Provide safe defaults and next steps when data is missing.

Explore prompting guides

Gateway recipes

Core gateway recipes

End-to-end request and streaming examples for the most common gateway workloads.

Chat completion recipe

Standard request with full JSON response.

curl https://<gateway-host>/v1/chat/completions \
  -H "Authorization: Bearer $DISRUPTIVERAIN_CLIENT_ID:$DISRUPTIVERAIN_CLIENT_SECRET" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {"text": "Summarize the top three risks from the report."},
    "stream": false
  }'

Streaming chat recipe

Receive SSE chunks for low-latency UI updates.

curl -N https://<gateway-host>/v1/chat/completions \
  -H "Authorization: Bearer $DISRUPTIVERAIN_CLIENT_ID:$DISRUPTIVERAIN_CLIENT_SECRET" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {"text": "Stream this response."},
    "stream": true
  }'

Audio transcription

Upload WAV or stream PCM frames over WebSocket.

curl https://<gateway-host>/v1/audio/transcriptions \
  -H "Authorization: Bearer $DISRUPTIVERAIN_CLIENT_ID:$DISRUPTIVERAIN_CLIENT_SECRET" \
  -F file=@sample.wav

Text-to-speech

Generate speech with voice, emotion, and speed controls.

curl https://<gateway-host>/v1/audio/speech \
  -H "Authorization: Bearer $DISRUPTIVERAIN_CLIENT_ID:$DISRUPTIVERAIN_CLIENT_SECRET" \
  -H "Content-Type: application/json" \
  -d '{"text":"Welcome aboard","voiceId":"default","emotion":"neutral"}'

Avatar streaming

Receive synchronized audio and blendshape frames.

const ws = new WebSocket('wss://<gateway-host>/v1/avatar/stream');

ws.onopen = () => {
  ws.send(JSON.stringify({
    text: 'Hello from the avatar stream',
    sessionId: 'avatar_123',
    voiceId: 'default',
    emotion: 'neutral',
    speed: 1.0,
  }));
};

ws.onmessage = (event) => {
  const message = JSON.parse(event.data);
  if (message.type === 'audio') console.log(message.audio);
  if (message.type === 'blendshape') console.log(message.frameIndex);
};

Fine-tuning jobs

Create a fine-tuning job and monitor progress.

curl https://<gateway-host>/v1/fine-tuning/jobs \
  -H "Authorization: Bearer $DISRUPTIVERAIN_CLIENT_ID:$DISRUPTIVERAIN_CLIENT_SECRET" \
  -H "Content-Type: application/json" \
  -d '{
    "modelId": "sidis-1",
    "trainingData": {"type": "file", "id": "file_123"},
    "hyperparameters": {"epochs": 3, "batchSize": 8}
  }'

Agent workflows

Agentic playbooks for production

Reference architectures for tool routing, memory, multimodal streams, and eval loops.

Support triage agent

Routes inbound tickets, calls retrieval for account context, and escalates with a crisp summary.

RoutingRAGHandoff

Tool-using research agent

Chains search, file analysis, and summarization into a single multi-tool runbook.

ToolsPlanningCitations

Realtime voice concierge

Streams audio input, turns it into actions, and speaks back with low latency.

AudioStreamingLatency

Eval flywheel

Captures failures, generates test cases, and tracks prompt regressions over time.

EvalsTelemetryQuality

Code review assistant

Reads diffs, flags risks, and outputs structured review checklists.

CodeGuardrailsChecks

Topic hubs

Explore by topic

Jump into the cookbook areas that map to your system design and product roadmap.

Prompting systems

Design prompt contracts that control tone, format, tool usage, and fallbacks across the Gateway API.

Prompt contracts
Few-shot patterns
Prompt testing

Agents + tools

Plan agent workflows that orchestrate tools, manage state, and hand off to humans when needed.

Tool routing
State + memory
Handoffs

Multimodal

Build multimodal systems that blend audio, vision, and avatar streams while keeping latency low.

Speech in/out
Vision analysis
Avatar streaming

Text workflows

Ship text systems that handle summarization, extraction, classification, and report generation at scale.

Summaries
Extraction
Classification

Guardrails

Ship safer agents with policy-first prompts, red-team testing, and structured evaluation loops.

Policy rules
Red-team prompts
Risk scoring

Optimization

Tune token budgets, caching, and concurrency to deliver reliable performance at scale.

Caching
Token budgets
Concurrency

Retrieval + memory

Use retrieval, vector search, and memory notes to ground responses in trusted data.

Vector search
Chunking
Citations

Open models

Deploy open models with vLLM, LoRA adapters, parity checks, and eval coverage.

vLLM runtime
LoRA adapters
Eval parity

Realtime + streaming

Build real time streaming apps with SSE, WebSocket, and event-driven UI updates.

Event schemas
Backpressure
Reconnects

Featured cookbooks

Featured guides and playbooks

Deep dives into the Gateway API, ADK patterns, and SIDIS orchestration for production-ready AI systems.

PromptingAgentsReasoning

Agentic prompting guide

Prompting patterns for agentic predictability, instruction adherence, and tool use.

Intermediate18 min read

Agents SDKToolsWorkflows

Multi-tool orchestration with the ADK

Plan, call tools, and summarize into a single runbook.

Intermediate18 min read

AudioRealtimeStreaming

Realtime voice assistant

Stream audio in, synthesize audio out, and handle turn taking.

Intermediate16 min read

ResponsesSearchCitations

RAG on PDFs with file search

Chunk documents, search with citations, and answer precisely.

Intermediate20 min read

EvalsQualityAutomation

Eval flywheel for prompt regressions

Generate test cases, score outputs, and track regressions.

Advanced14 min read

Open ModelsOpsDeployment

Self-hosted model deployment

Run open models locally with parity checks and cost controls.

Advanced16 min read

AvatarStreamingMultimodal

Avatar streaming guide

Build real-time avatar experiences with synchronized audio and facial animation.

Advanced20 min read

AgentsWorkflowsADK

Multi-agent workflow patterns

Coordinate complex workflows across specialized agents with governance and handoffs.

Advanced22 min read

Popular cookbooks

The patterns builders reach for the most.

Function calling guide

Functions

Define tool schemas, validate arguments, and handle retries safely.

Policy-first prompting

Safety

Layer safety policy before task instructions to reduce risk.

Prompt caching 101

Latency

Reduce latency and cost with cache-safe prompt blocks.

Streaming formats and reconnects

Streaming

Event schemas, heartbeats, and reconnect logic for SSE and WebSocket.

Summarizing long documents

Completions

Chunk, summarize, and merge outputs with quality checks.

Latest updates

Fresh from the lab

New recipes and updates across prompting, agents, and multimodal workflows.

UpdatedEvals

Eval flywheel for prompt regressions

Added structured output scoring and drift dashboards.

NewReasoning

Reasoning transparency patterns

New guide on reasoning display patterns.

NewAgents

Multi-agent workflow patterns

Complete orchestration patterns for enterprise workflows.

NewConnectors

Connector integrations guide

Complete connector setup and OAuth flow guide.

NewGateway

Gateway API authentication guide

Complete authentication patterns for Gateway API.

NewAvatar

Avatar streaming guide

Complete guide for avatar API integration.

UpdatedStreaming

Streaming formats and reconnects

Updated reconnect examples and backpressure handling.

UpdatedOpen Models

Self-hosted model deployment

Added parity tests for self-hosted rollouts.

Launch checklist

Production readiness

Operational guardrails to keep your gateway workloads resilient in production.

Backoff on rate limits

Honor Retry-After headers and cap concurrent streams for audio workloads.

Telemetry on every request

Log token usage, latency, tool calls, and completion status.

Prompt regression tests

Run eval suites on every prompt or model change.

Schema validation

Validate JSON outputs and retry with correction prompts.

Key rotation + scopes

Use per-environment keys and least-privilege scopes.

Human escalation path

Define when agents should hand off to an operator.

Ready for production?

Pair these recipes with eval suites, usage tracking, and environment-specific API keys before launch.

Return to API reference