# Doubleword Documentation

> Docs and references for the Doubleword inference platform with realtime, async, and batch inference and the open-source Control Layer gateway.

All documentation pages are available as markdown by appending `.md` to any URL.

## CLI

Full documentation: [CLI](/dw-cli.md)

Command-line interface for the Doubleword Inference API and Platform.

### Doubleword CLI

- [Introduction](/dw-cli/introduction.md)

### Getting Started

- [Installation](/dw-cli/installation.md)
- [Authentication](/dw-cli/authentication.md)
- [Quickstart](/dw-cli/quickstart.md)

### Core Workflows

- [Batch Processing](/dw-cli/batches.md)
- [Streaming Results](/dw-cli/streaming.md)
- [Real-Time Inference](/dw-cli/realtime.md)

### File Tools

- [JSONL Format](/dw-cli/jsonl-format.md)
- [Local File Operations](/dw-cli/file-tools.md)

### Projects

- [Project System](/dw-cli/projects.md)
- [Examples](/dw-cli/examples.md)

### Reference

- [Command Reference](/dw-cli/commands.md)
- [Accounts & Configuration](/dw-cli/accounts.md)
- [API Keys & Webhooks](/dw-cli/keys-webhooks.md)
- [Usage & Analytics](/dw-cli/usage.md)
- [Global Flags](/dw-cli/global-flags.md)

## Control Layer

Full documentation: [Control Layer](/control-layer.md)

The Control Layer is a distributed AI model management platform that provides secure, scalable access control between your users and AI model providers.

### Getting Started

- [Control Layer Overview](/control-layer/control-layer-overview.md): Overview of the Doubleword Control Layer, a platform for AI model management with secure and scalable access control.
- [Getting Started](/control-layer/getting-started.md)

### How-to Guides

- [Connect to the API](/control-layer/how-to/api.md): Learn how to interact with AI models using the Control Layer's OpenAI-compatible API, including configuration, request handling, and best practices.
- [Add Endpoints](/control-layer/how-to/endpoints.md): Learn how to connect the Control Layer to AI inference endpoints by configuring model sources.
- [Manage Users & Groups](/control-layer/how-to/users-and-groups.md): Learn how to manage user and group access in Control Layer for efficient model permissions.
- [Set Up Health Monitoring](/control-layer/how-to/health-monitoring.md): Learn how to monitor the status of AI model endpoints using Control Layer's built-in health monitoring features.
- [Deploy to Production](/control-layer/how-to/deploy-to-production.md): Checklist for deploying Control Layer to a production environment.
- [Set Up Model Pricing](/control-layer/how-to/tariffs.md): Configure per-token pricing for your models using tariffs. Set different rates for realtime, batch, and playground usage.
- [Set Up Payments](/control-layer/how-to/payments.md): Enable credit purchases via Stripe so users can buy credits through a self-service checkout flow.

### Reference

- [Configuration Reference](/control-layer/reference/configuration.md): Reference guide for configuring the Control Layer using YAML and environment variables.

### Conceptual Guides

- [How Billing Works](/control-layer/conceptual-guides/how-billing-works.md): Understand the credits-based billing system: how charges are calculated, what happens at zero balance, and how payments work.

## Doubleword Inference API

Full documentation: [Doubleword Inference API](/inference-api.md)

The Doubleword Inference API provides high performance, high throughput LLM inference at low cost. Sign up [now](https://app.doubleword.ai).

### Reference

- [API Reference](/inference-api/api-reference.md): Complete API reference for the Doubleword Batch API
- [API Skill](/inference-api/skill.md): Download a Claude Code skill that teaches Claude how to use the Doubleword Batch API for high-throughput workloads.

### Overview

- [Intro to Doubleword Inference](/inference-api/intro-to-doubleword-inference.md)
- [Realtime Inference](/inference-api/realtime-inference.md)
- [Async Inference](/inference-api/async-inference.md)
- [Batch Inference](/inference-api/batch-inference.md): A guide on how to submit your first batch to the Doubleword Batched API

### Models

- [Models & Pricing](/inference-api/models.md): Browse available Doubleword models and pricing.

### Using the Platform

- [Using the Platform](/inference-api/using-the-platform.md): Platform integration guides
- [What is a JSONL file?](/inference-api/jsonl-files.md): Understand JSONL formatting and see step-by-step examples for submitting large batches of chat completion requests to the Doubleword Batch API.
- [Creating an API Key](/inference-api/creating-an-api-key.md)
- [Prompt caching](/inference-api/prompt-caching.md): Reuse a large, unchanging prompt prefix across requests and pay a reduced rate for the repeated tokens using cache_control markers.
- [Tool Calling and Structured Outputs](/inference-api/tool-calling.md): Learn how to use OpenAI-compatible tool calling and Structured Outputs with Doubleword to build reliable agents and enforce strict JSON Schema responses.
- [Reasoning effort](/inference-api/reasoning-controls.md): Set reasoning effort with the OpenAI-compatible Chat Completions or Responses API.
- [Notifications & Webhooks](/inference-api/batch-notifications-and-webhooks.md)

### Account & Billing

- [Account & Billing](/inference-api/account-and-billing.md): Account and billing guides
- [Adding credits to your account](/inference-api/adding-credits-to-your-account.md): Step-by-step guide to adding credits in the Doubleword Console so your Batch API calls run without interruptions.
- [Auto Top-Up](/inference-api/auto-topup.md): Automatically recharge your credit balance when it drops below a threshold, so you never run out mid-workflow.
- [How to manage payment methods and invoice history](/inference-api/how-to-manage-payments.md)

### Organizations

- [Organizations Overview](/inference-api/organizations-overview.md)
- [Inviting Team Members](/inference-api/inviting-team-members.md)
- [Organization API Keys](/inference-api/organization-api-keys.md)
- [Submitting Batches as an Organization](/inference-api/organization-batches.md)
- [Organization Credits & Billing](/inference-api/organization-credits.md)

### SDKs & CLI

- [SDKs & CLI](/inference-api/sdks-and-cli.md): Client libraries and CLI docs
- [Autobatcher](/inference-api/autobatcher.md): Python and TypeScript client that automatically converts individual API calls into batched requests for cost savings.
- [DW CLI](/inference-api/dw-cli.md): Install and use the Doubleword CLI to manage batch inference from the terminal.

### Workbooks

- [Workbooks](/inference-api/cli-examples.md)
- [Async Agents](/inference-api/async-agents.md)
- [Data Processing Pipelines](/inference-api/data-processing-pipelines.md)
- [Structured Data Extraction](/inference-api/structured-extraction.md)
- [Large-Scale Semantic Search Without Embeddings](/inference-api/semantic-search-without-embeddings.md): Implement semantic search at scale using LLMs directly instead of traditional embedding models.
- [Creating a Daily Research Paper Digest](/inference-api/research-summaries.md): Build an automated arXiv digest that costs $1/year using the Batch API to filter and summarize relevant papers.
- [Image Summarization](/inference-api/image-summarization.md)
- [Embeddings](/inference-api/embeddings.md)
- [Model Evals](/inference-api/model-evals.md)
- [Synthetic Data Generation](/inference-api/synthetic-data-generation.md)
- [Dataset Compilation](/inference-api/dataset-compilation.md)
- [Classifying Security Vulnerabilities ](/inference-api/bug-detection-ensemble.md)
- [Building an Agentic PR Review Bot](/inference-api/pr-review-bot.md): Build a self-hosted GitHub PR review bot on opencode and Doubleword async inference, with code samples for chat completions and the Open Responses API.
- [Doubleword Agent Swarm](/inference-api/agent-swarm.md): Run a self-designing agent swarm — orchestrator, bounded-context workers, adversarial verifiers, synthesizer — on the Open Responses API with the dw CLI.

### Integrations

- [Integrations](/inference-api/integrations.md): Use the Doubleword Inference API with popular AI agent frameworks
- [Anthropic API Compatibility](/inference-api/anthropic-api-compatibility.md): Use the Anthropic Messages API against Doubleword - messages, streaming, the flex tier, and batch via the OpenAI batch format.
- [Agno](/inference-api/integrations/agno.md): Route Agno agent requests through the Control Layer
- [Arize AX](/inference-api/integrations/arize-ax.md): Evaluating agentic workflows gets expensive fast. Run your LLM-as-a-judge evals on Doubleword's high-throughput batch tier for 4-6x less and watch every generation, judgement, and score land in Arize AX.
- [Arize Phoenix](/inference-api/integrations/arize-phoenix.md): Integrate Self-Hosted Arize Phoenix with Doubleword to Run LLM-as-Judge Evals at Scale
- [atomic-agents](/inference-api/integrations/atomic-agents.md): Route atomic-agents requests through the Control Layer
- [CrewAI](/inference-api/integrations/crewai.md): Route CrewAI agent requests through the Control Layer
- [Daytona](/inference-api/integrations/daytona.md): Offload your agent's code to the background with Doubleword and Daytona, generating it cheaply with async inference and running it safely in parallel.
- [Google ADK](/inference-api/integrations/google-adk.md): Route Google ADK agent requests through the Control Layer
- [LangChain / LangGraph](/inference-api/integrations/langchain.md)
- [Langsmith](/inference-api/integrations/langsmith.md)
- [LlamaIndex](/inference-api/integrations/llamaindex.md)
- [Mastra](/inference-api/integrations/mastra.md): Route Mastra agent requests through the Control Layer
- [mem0](/inference-api/integrations/mem0.md): Route Mem0's memory extraction through Doubleword's async Flex tier to bulk-ingest thousands of transcripts for half the cost, with no rate-limit backoff code.
- [Microsoft Agent Framework](/inference-api/integrations/microsoft-agent-framework.md): Route Microsoft Agent Framework requests through the Doubleword API
- [OpenAI Agents SDK](/inference-api/integrations/openai-agents.md): Route OpenAI Agents SDK requests through the Control Layer
- [OpenClaw](/inference-api/integrations/openclaw.md)
- [Opencode](/inference-api/integrations/opencode.md): Run the most performant open-source coding model in your terminal or local browser. This guide walks you through pairing opencode with GLM-5.2 on Doubleword - full 1M-token context, ideal for long-running background tasks and agents, and repository-wide work.
- [PydanticAI](/inference-api/integrations/pydantic-ai.md): Route PydanticAI agent requests through the Control Layer
- [smolagents](/inference-api/integrations/smolagents.md): Route smolagents requests through the Control Layer
- [Supermemory](/inference-api/integrations/supermemory.md): Learn how to cut Supermemory memory extraction costs by 99% by routing background LLM workloads through Doubleword's async and batch API tiers, while keeping real-time retrieval fast.
- [Vercel AI SDK](/inference-api/integrations/vercel-ai.md)

### Data Sources

- [Data Sources](/inference-api/data-sources.md): Connect external data sources to Doubleword for automated batch processing.
- [S3 Connections (Beta)](/inference-api/data-sources/connections.md): Connect S3 buckets to automatically sync files and create batches.

### Support

- [Get Support](/inference-api/get-support.md): Contact support, request features, or get help optimizing your batch workloads.

### Archive

- [Why Batch Inference Matters](/inference-api/why-batch-inference-matters.md): Understand how batch processing enables autonomous AI workloads that deliver higher business value than real-time assistants.
- [Behind the Stack: Batched Endpoints](/inference-api/behind-the-stack-batched-endpoints.md): Deep dive into how batched endpoints reduce costs through demand flattening, spot instances, and prefix caching.
- [Parallel Primitives for Multi-Agent Workflows](/inference-api/parallel-primitives.md): Design patterns for running multi-agent workflows efficiently using parallel batch processing.
- [ZeroDP: Just-In-Time Weight Offloading over NVLink for Data Parallelism](/inference-api/zerodp-just-in-time-weight-offloading-over-nvlink-for-data-parallelism.md): High throughput inference of LLMs using JIT weight offloading to optimize KV Cache.

### Models

- [GLM 5.2](/inference-api/models/zai-org-glm-5-2-fp8.md)
- [Nemotron 3 Ultra 550B A55B](/inference-api/models/nvidia-nvidia-nemotron-3-ultra-550b-a55b-nvfp4.md)
- [DeepSeek V4 Pro](/inference-api/models/deepseek-ai-deepseek-v4-pro.md)
- [DeepSeek V4 Flash](/inference-api/models/deepseek-ai-deepseek-v4-flash.md)
- [Qwen3.6 35B A3B](/inference-api/models/qwen-qwen3-6-35b-a3b-fp8.md)
- [Kimi K2.6](/inference-api/models/moonshotai-kimi-k2-6.md)
- [GLM 5.1](/inference-api/models/zai-org-glm-5-1-fp8.md)
- [Gemma 4 31B IT](/inference-api/models/google-gemma-4-31b-it.md)
- [Nemotron 3 Super 120B A12B](/inference-api/models/nvidia-nvidia-nemotron-3-super-120b-a12b-nvfp4.md)
- [Qwen3.5 9B dottxt](/inference-api/models/qwen-qwen3-5-9b-dotjson.md)
- [Qwen3.5 4B](/inference-api/models/qwen-qwen3-5-4b.md)
- [Qwen3.5 9B](/inference-api/models/qwen-qwen3-5-9b.md)
- [Qwen3.5 35B A3B dottxt](/inference-api/models/qwen-qwen3-5-35b-a3b-fp8-dotjson.md)
- [Qwen3.5 397B A17B dottxt](/inference-api/models/qwen-qwen3-5-397b-a17b-fp8-dottxt.md)
- [Qwen3.5 397B A17B](/inference-api/models/qwen-qwen3-5-397b-a17b.md)
- [DeepSeek OCR 2](/inference-api/models/deepseek-ai-deepseek-ocr-2.md)
- [LightOnOCR 2 1B bbox soup](/inference-api/models/lightonai-lightonocr-2-1b-bbox-soup.md)
- [Devstral 2 123B Instruct 2512](/inference-api/models/mistralai-devstral-2-123b-instruct-2512.md)
- [olmOCR 2 7B 1025](/inference-api/models/allenai-olmocr-2-7b-1025-fp8.md)
- [Qwen3 VL 30B A3B Instruct](/inference-api/models/qwen-qwen3-vl-30b-a3b-instruct-fp8.md)
- [Qwen3 VL 235B A22B Instruct](/inference-api/models/qwen-qwen3-vl-235b-a22b-instruct-fp8.md)
- [GPT OSS 20B](/inference-api/models/openai-gpt-oss-20b.md)
- [GPT OSS 120B](/inference-api/models/openai-gpt-oss-120b.md)
- [Qwen3 Embedding 8B](/inference-api/models/qwen-qwen3-embedding-8b.md)
- [Qwen3 14B](/inference-api/models/qwen-qwen3-14b-fp8.md)