Skip to main content

​
Retrievers

Retrievers are responsible for finding relevant information from a configured Knowledge source. In BindAI V1, retrieval is part of the project-local Knowledge runtime. An agent can be configured with a Knowledge source, and when the agent executes, BindAI searches that source and makes relevant content available to the agent. The V1 retrieval path is intentionally simple:
BindAI V1 does not expose separate vector, BM25, hybrid, or reranking configuration in Studio.

​
V1 Retrieval Flow

A Knowledge-enabled agent can use project Knowledge during execution. The basic flow is:
Knowledge retrieval is therefore different from the model call itself. The model generates the response, while the Knowledge runtime provides information from the configured project Knowledge source.

​
Configuring Knowledge for an Agent

Knowledge is associated with an agent through its Knowledge configuration. For example, a project can contain a Knowledge source named:
An agent can then reference that Knowledge source. Conceptually:
During execution, BindAI resolves the configured Knowledge source and searches it when Knowledge is used by the agent. The Studio interface exposes Knowledge configuration for agents.

​
Project Knowledge

BindAI V1 uses project-local Knowledge. Knowledge belongs to the project rather than being a global retrieval service. A simplified project structure can look like:
The Knowledge runtime searches the configured project Knowledge source. This keeps Knowledge close to the project that uses it and allows the same Knowledge configuration to be used by the project’s agents.

​
Search Results

The important V1 retrieval contract is the resulting Knowledge content. For example, a Knowledge document may contain:
A Knowledge-enabled agent can search the configured source and retrieve that information during execution. The application does not need to manually load the document into every agent prompt.

​
Concrete V1 Example

Consider a project with:
The document contains:
An agent is configured to use:

​
Run 1

The workflow executes the Knowledge-enabled agent with a request that requires information from the configured Knowledge source. The Knowledge runtime searches knowledge-test and can retrieve:
The retrieved information is then available during the agent execution.

​
Run 2

The same workflow is executed again with another request that requires information from the same Knowledge source. BindAI again resolves:
The agent can retrieve the same project Knowledge without the application manually loading knowledge-test.md into the workflow. The important point is that Knowledge retrieval happens as part of the configured agent runtime rather than requiring the workflow author to hard-code the document contents.

​
Retrieval and Agents

Knowledge retrieval is directly connected to agent execution in V1. The runtime relationship is:
This allows an Agent node in a workflow to use project Knowledge while remaining a normal workflow node.

​
Retrieval and Workflows

Retrieval is not a separate workflow node in V1. Instead, Knowledge is configured on the agent that uses it. For example:
The workflow controls execution order, while the agent’s Knowledge configuration controls which project Knowledge is available during that agent execution.

​
Retrieval and Tools

Retrieval and Tools have different responsibilities.
For example, an agent can use Knowledge to find reference information and then use a Tool to perform an operation based on that information. Connection-backed Webhook tools are also available in V1, but they are execution tools rather than retrievers.

​
Retrieval and Memory

Knowledge retrieval and Memory solve different problems.
Memory in V1 is used for persisted conversation context. Knowledge is used to search project reference content. An agent can use both:
They should not be treated as interchangeable storage systems.

​
Retrieval and Workflow State

Retrieved Knowledge should also be distinguished from workflow state. Workflow state contains values produced while the workflow executes. Knowledge is project reference information that the agent can search. Conceptually:
This separation allows a workflow to execute normally while its agents use Knowledge as an external reference source within the project.

​
Knowledge Search in Studio

Studio provides the V1 Knowledge interface for project Knowledge. The implemented flow includes:
  1. Create or manage project Knowledge.
  2. Add Knowledge content to the project.
  3. Configure an agent to use a Knowledge source.
  4. Execute the agent or workflow.
  5. Allow the runtime to search the configured Knowledge during execution.
The Knowledge configuration can be inspected from the Studio agent configuration interface.

​
Retrieval Testing

A useful V1 retrieval test should use a known value in a project Knowledge document. For example:
Then execute an agent configured with that Knowledge source using a request that requires the stored information. The test should verify that the agent execution can access the known Knowledge content. A second execution can verify that the same configured Knowledge source remains available across separate workflow runs. This tests the actual V1 integration:

​
What V1 Does Not Expose

The current BindAI V1 Studio does not expose separate configuration for:
  • Vector retrieval
  • BM25 retrieval
  • Hybrid retrieval
  • Embedding-based search configuration
  • Metadata-filtered retrieval
  • Reranking
  • Lexical reranking
  • Conversational retrieval configuration
  • Top-K retrieval controls
  • Vector databases
  • External search indexes
These concepts should not be assumed to be available simply because they are common in RAG systems. V1 documents the retrieval behavior that is actually available through the project Knowledge runtime.

​
Retrieval Security

Knowledge can contain project-specific information. Applications should therefore treat Knowledge content as project data and ensure that the project and its API access are appropriately protected. Knowledge retrieval does not replace application authentication or authorization. The Studio API uses the configured BindAI API authentication mechanism when accessing project resources.

​
Retrieval vs. RAG

Retrieval is one part of a broader RAG-style architecture, but BindAI V1 does not expose a separate configurable RAG pipeline. The implemented path is closer to:
Therefore, V1 documentation should focus on the actual Knowledge-to-Agent integration rather than describing an unimplemented vector RAG pipeline.

​
Current V1 Retrieval Capabilities

BindAI V1 provides:
  • Project-local Knowledge sources
  • Knowledge content used during agent execution
  • Agent Knowledge configuration
  • Runtime Knowledge search
  • Studio Knowledge management
  • Knowledge use inside workflow Agent nodes
  • Repeated Knowledge retrieval across workflow runs
The V1 retrieval implementation is intentionally narrower than a full vector-search platform.

​
Summary

Retrievers in BindAI V1 provide the search step between configured project Knowledge and an agent. The implemented architecture is:
Knowledge retrieval works with the existing Agent, Workflow, Memory, Tool, and Studio systems. V1 does not expose separate vector, BM25, hybrid, reranking, or external retrieval configuration. The supported retrieval behavior is the project-local Knowledge search used by configured agents.