Haystack
Haystack is an open-source framework by deepset for language-model applications in Python. Modular pipelines and agent workflows explicitly define how information is retrieved, filtered and passed to the model.
- Project
- Haystack
- Publisher
- deepset
- Repository
- github.com/deepset-ai/haystack
- Documentation
- docs.haystack.deepset.ai/docs/intro
- Languages
- Python
- Category
- AI agent frameworks
What is Haystack?
Haystack is an orchestration framework by deepset, released under the Apache 2.0 licence. Applications are assembled from components into pipelines – for RAG, semantic search, question answering, multimodal applications and agents. The project centres on context engineering: retrieval, ranking, filtering and routing of information remain visible and traceable.
Core capabilities
- Pipelines
- Components for retrieval, indexing, tool calling, memory and evaluation, connected with loops, branches and conditions.
- Agents
- Lifecycle hooks for guardrails and custom logic; steps, token usage and tool calls are tracked for monitoring and cost control.
- Asynchronous execution
- A pipeline runs synchronously or asynchronously and can stream token by token.
- Vendor independence
- Integrations with OpenAI, Mistral, Anthropic, Cohere, Hugging Face, Google, Azure OpenAI, AWS Bedrock and local models, among others.
- Deployment
- The companion project Hayhooks exposes pipelines and agents as REST APIs or MCP servers.
When Haystack fits
Haystack fits when you want to define precisely which context reaches a model – for search and answers over large document collections, for example. Note: Haystack collects anonymous usage statistics on pipeline components by default; the documentation explains how to opt out.
Using Haystack effectively in an AI-first business model
Haystack’s explicit structure supports a requirement that is often underestimated in operation: being able to trace why an agent gave a particular answer or proposed an action.
Use hooks and counters to implement guardrails and cost limits that you have agreed in the operating model beforehand. Which limit applies and who decides when it is exceeded is a business question – Haystack provides the points where it takes effect technically.
From tool to outcome
In AI-first coaching you clarify which outcomes AI agents should be accountable for in your go-to-market and within which framework. REVAGO then designs, builds and runs your system. You pay for results – not for activity.
Frequently asked questions
What is Haystack?
Haystack is an open-source framework by deepset (Apache 2.0 licence) for language-model applications in Python. It assembles pipelines and agent workflows from components and is suited to RAG, semantic search and agents, among other uses.
Can Haystack build agents?
Yes. Haystack includes an agent component with tool calling, concurrent tool calls, lifecycle hooks for guardrails and counters for steps and token usage.
How do I deploy a Haystack pipeline?
The companion project Hayhooks exposes pipelines and agents as HTTP endpoints or MCP servers, including OpenAI-compatible chat endpoints. Docker images are also available.
Further reading in the wiki
A neutral overview, not a recommendation of individual providers. REVAGO is not affiliated with the listed projects. Descriptions are based on the projects’ public information; features and licences may change.