Tool: DSPy — Zelaxy Docs

Execute DSPy language-model programs hosted on your own server and receive structured predictions. Use this block when you want to run compiled, optimized DSPy pipelines — such as chain-of-thought modules, multi-hop reasoners, or fine-tuned prompt chains — from inside a Zelaxy workflow.

Overview

Property	Value
Type	`dspy`
Category	Tool — AI & ML
Auth	API Key (Bearer token, optional)

Operations

Operation	Tool ID	Description
DSPy Run	`dspy_run`	POST a DSPy program and an input string to your server's `/run` endpoint and return the program's output and full prediction object.

Configuration

Setting	Type	Required	Description
`baseUrl`	`string`	Yes	Base URL of your DSPy server (e.g., `https://your-dspy-server.com`). Trailing slashes are stripped automatically.
`apiKey`	`string`	No	Bearer token for server authentication. Set this if your DSPy server requires an API key. Store as a secret and reference with `{{DSPY_API_KEY}}`.
`program`	`string`	Yes	Python DSPy program code to execute. The block's code editor lets you write or paste the full module definition.
`input`	`string`	Yes	Input value passed to the DSPy program at runtime. Can come from an upstream block using `{{blockName.output}}`.

Outputs

Field	Type	Description
`output`	`string`	Primary text result returned by the DSPy program — resolved from the server response's `output`, `answer`, or `response` field (in that priority order).
`prediction`	`json`	Full prediction object returned by the server. Contains all intermediate fields produced by the DSPy module (e.g., rationale, intermediate steps).

Example

[Starter] → [DSPy: DSPy Run] → [Agent: use the result]

A Starter block collects a user question and passes it as {{starter.input}}. The DSPy block connects to {{DSPY_BASE_URL}}, authenticates with {{DSPY_API_KEY}}, runs a chain-of-thought program, and surfaces {{dspy.output}} to a downstream Agent that formats the final response for the user.

Tips

Self-hosted server required — DSPy programs run on your own inference server, not on a third-party API. Deploy a FastAPI or Flask wrapper that exposes a POST /run endpoint accepting { program, input } and returning { output, prediction }.
API key is optional — if your server is behind a private network or has no authentication, leave apiKey blank. If it does require auth, add the key as an environment secret and reference it as {{DSPY_API_KEY}}.
prediction for debugging — the raw prediction JSON output contains every intermediate field DSPy produced (e.g., rationale traces). Pass it to an Agent or log it in a Response block while iterating on your program.
Upstream inputs — pipe dynamic content into the input field with {{blockName.field}} syntax so the same DSPy program can handle varying user data without changing the program code.

DSPy