# Dust Dust is a developer platform for building AI agents and custom workflows, launched in 2023 by a team led by former OpenAI engineers. It offers a suite of tools to design, test, and deploy AI assistants across enterprises, with a focus on custom models and data integration. Dust is a developer platform for designing, building, and deploying AI agents and custom workflows, targeting enterprises and engineering teams. Founded in 2023, the company emerged from the growing ecosystem around [large-language-model](https://www.wikiprompt.org/wiki/large-language-model)s, aiming to provide infrastructure for agentic applications that go beyond simple chat interfaces. Dust's core offering includes a drag-and-drop visual builder, a code-first SDK, and a managed runtime, enabling teams to create, test, and monitor AI assistants that interact with external tools, APIs, and internal databases. The platform emerged from a period of rapid advancement in [generative-ai](https://www.wikiprompt.org/wiki/generative-ai) and [transformer](https://www.wikiprompt.org/wiki/transformer) architectures, addressing a need for production-grade tooling. Dust's approach emphasizes flexibility and control, allowing developers to plug in any [large language model](https://www.wikiprompt.org/wiki/large-language-model) from providers such as [OpenAI](https://www.wikiprompt.org/wiki/openai), [Anthropic](https://www.wikiprompt.org/wiki/anthropic), or [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind), or use custom models fine-tuned on private datasets. This positions Dust against no-code agent builders and more generalized orchestration frameworks, focusing on scalability, observability, and collaboration for software teams. ## Founding and Early History Dust was co-founded in 2023 by Francois-Xavier Caulier, Gabriel Hubert, and Stanislas Polu, three engineers who previously worked at [OpenAI](https://www.wikiprompt.org/wiki/openai). Their experience with cutting-edge AI research and deployment informed Dust's initial design: a platform to streamline the transition from prototype to production. The first public beta launched in early 2024, quickly gaining traction in the developer community for its ease of use and deep technical integrations. Initial funding rounds included a $2 million seed from notable investors including Sequoia Capital, followed by a $16 million Series A in February 2025 led by Greylock Partners, pushing the company to a $100 million valuation. By mid-2024, Dust had built its customer base to over 200 companies, including a mix of startups and Fortune 500 firms, with particular adoption in the financial services and healthcare sectors. The platform saw a major update in October 2025 with its 2.0 release, introducing a versioned API, a deterministic (non-probabilistic) execution engine for [machine learning](https://www.wikiprompt.org/wiki/machine-learning) pipelines, and integration with [Amazon Web Services](https://www.wikiprompt.org/wiki/amazon-web-services) and [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud) for on-premise or cloud-centric deployments. ## Key Features and Products The primary product is the Dust console, which offers several distinct components: - **Visual Task Builder**: A browser-based drag-and-drop canvas for creating workflow sequences, connecting LLMs, memory blocks, and tool calls. The builder generatesTypeScript code automatically, which can be committed directly to a team's version-control system. - **Code IDE**: For advanced users, a YAML and TypeScript library allows hand-coding workflows, with syntax highlighting and a local testing harness. The SDK supports social coding through comments and version history. - **Model Gateway**: A unified API to access LLMs from [OpenAI](https://www.wikiprompt.org/wiki/openai), [Anthropic](https://www.wikiprompt.org/wiki/anthropic), [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind), and [Meta AI](https://www.wikiprompt.org/wiki/meta-ai), plus support for open-weights models like Llama 3 via [CoreWeave](https://www.wikiprompt.org/wiki/coreweave) or [Groq](https://www.wikiprompt.org/wiki/groq) for low-latency inference. The gateway provides fallback routing and cost tracking. - **Memory and Retrieval**: Built-in vector storage for long-term memory and RAG (retrieval-augmented generation), supporting connectivity to various sources like Salesforce, Slack, and Notion. - **Observability**: Logging, token auditing, and latency tracking across the entire stack, from LLM calls to tool invocations, with custom dashboards. - **Evaluation Suite**: A suite to run unit tests and regression tests on agent performance against expected outputs. ## Technology At its core, Dust implements a chain-of-thought reasoning framework, but it distinguishes itself by allowing users to break sentences explicitly none. The platform abstracts away the complexities of prompt engineering and token management, offering a domain-specific language (DSL) that compiles into executable graph consisting of nodes for LLM calls, tools, or decision branches. This architecture enables a hybrid localization - the entire pipeline can run on dedicated cloud infrastructure using **orchestration**, or a developer's own VPC for data privacy. Dust has implemented a proprietary routing algorithm that predicts accuracy-cost tradeoffs between different model providers, which up to a 30% cost reduction in large production workloads, according to internal benchmarks published in January 2025. For inference, Dust supports both synchronous on-demand calls and batch processing. Its runtime scheduler optimizes GPU usage by batching concurrent requests from multiple workflows, achieving throughput of up to 200 requests per second on a single p3.16xlarge instance (using [[AWS Tr] ## Enterprise Adoption and Use Cases Businesses use Dust.txt primarily for internal AI assistants specialized to handle tasks like ticket routing, code review, and document summarization. For instance: - **Financial Services**: A investment bank deployed Dust to automate its research digest processes, ingesting over 10,000 Reuters and Bloomberg feeds daily and producing personalized briefings in under 15 minutes. - **Healthcare**: One digital health startup used Dust to build an AI patient-triage agent that integrates with their EHR system, handling 2,500 conversations per day with a reported 15% escalation rate. The compliance built into the harness ensures HIIPA compliance with logging. - **Enterprise IT**: Another tech firm uses Dust to manage internal helpdesk, making ticketing workflows, achieving sustained first-response accuracy of, and improved labeled data. The platform's flexibility has also attracted independent developers, who use Dust on the free tier to build side projects, hobby agents, and educational tools. ## Community and Ecosystem Dust maintains an open-source contributions and has published research on its agent infrastructure, including a paper presented at ICML 2024 on "Robust Multi-Step Inference", co-authored with Stanford University's Artificial Intelligence Lab. Its blog has tutorials and runbooks. In 2025, the company started hosting annual developer conference, DustCon, attracting 1,200 attendees, and launched a partner incubator with $100,000 grants in seed funding for projects built on the platform. The broader ecosystem includes integrations with monitoring tools like [Langfuse](https://www.wikiprompt.org/wiki/langfuse) and [Weights & Biases](https://www.wikiprompt.org/wiki/weights-biases) and deployment platforms including Vercel. As a key contributor to the [Python, TypeScript] ecosystems, Dust distributes its SDK through PyPI and npm. ## Future Directions and Market Position In 2025, Dust expanded into the European market with a Paris office, adding over 50 employees, and announced a roadmap to incorporate more advanced efficiency techniques like speculative decoding and multi-criteria optimization. Executives emphasize that the biggest differentiator is not the model quality, but the ability to separate the orchestration from the actors. As of late 2025, Dust remains a relatively small startup, but its proactive adoption and focus on developer/solve positioning it favorably against forerunners in the agent-development space. Dust faces competition from established cloud giants like [Amazon Web Services](https://www.wikiprompt.org/wiki/amazon-web-services) and [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud) as they add native agent tooling, as well as from startups like LangChain and LlamaIndex. However, its focus on enterprise-grade reliability and its open-ended model connectivity gives it a unique angle in the market. ## See Also - [Generative AI](https://www.wikiprompt.org/wiki/generative-ai) - [Large Language Model](https://www.wikiprompt.org/wiki/large-language-model) - [AI agent](https://www.wikiprompt.org/wiki/ai-agent) --- Source: https://www.wikiprompt.org/wiki/dust-ai License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/) Last updated: 2026-10-07T16:32:37.672481+00:00