Prompt Management

Oodle includes a Langfuse-compatible prompt management API. You can use the standard Langfuse SDK to create, version, label, and fetch prompts at runtime — no separate Langfuse deployment required.

Prompts are managed in the Oodle UI at GenAI > Prompts (ap1, us1), where you can edit content, assign labels like production or staging, compare versions, and test in the playground.

Getting Started

1. Install the Langfuse SDK

Python

pip install langfuse

JavaScript / TypeScript

npm install langfuse

2. Configure credentials

The SDK needs three values:

Variable	Description
LANGFUSE_SECRET_KEY	Your Oodle API key (from Settings > API Keys ap1, us1)
LANGFUSE_HOST	https://<your-domain>/v1/api/instance/<instance>/langfuse
LANGFUSE_PUBLIC_KEY	Set to "default"

Set them as environment variables:

export LANGFUSE_HOST="https://<your-domain>/v1/api/instance/<instance>/langfuse"
export LANGFUSE_SECRET_KEY="<YOUR_API_KEY>"
export LANGFUSE_PUBLIC_KEY="default"

Or pass them when initializing the client:

Python

from langfuse import Langfuse

langfuse = Langfuse(
    public_key="default",
    secret_key="<YOUR_API_KEY>",
    host="https://<your-domain>/v1/api/instance/<instance>/langfuse",
)

JavaScript / TypeScript

import Langfuse from "langfuse";

const langfuse = new Langfuse({
  publicKey: "default",
  secretKey: "<YOUR_API_KEY>",
  baseUrl: "https://<your-domain>/v1/api/instance/<instance>/langfuse",
});

3. Create a prompt

Use {{variable}} syntax for template variables that will be filled in at runtime:

Python

langfuse.create_prompt(
    name="my-prompt",
    prompt="You are a helpful assistant.\n{{user_input}}",
    labels=["production"],
)

JavaScript / TypeScript

await langfuse.createPrompt({
  name: "my-prompt",
  prompt: "You are a helpful assistant.\n{{user_input}}",
  labels: ["production"],
});

Each call creates a new version. The latest label is automatically assigned to the newest version.

4. Fetch and compile a prompt

Fetch a prompt by name and label, then compile it with variable values:

Python

prompt = langfuse.get_prompt(
    "my-prompt",
    label="production",
)

compiled = prompt.compile(
    user_input="What is observability?",
)

JavaScript / TypeScript

const prompt = await langfuse.getPrompt(
  "my-prompt",
  { label: "production" },
);

const compiled = prompt.compile({
  user_input: "What is observability?",
});

You can also fetch by specific version number:

Python

prompt = langfuse.get_prompt(
    "my-prompt",
    version=3,
)

JavaScript / TypeScript

const prompt = await langfuse.getPrompt(
  "my-prompt",
  { version: 3 },
);

5. Verify in Oodle

Navigate to GenAI > Prompts (ap1, us1) to see your prompts, version history, and label assignments.

Labels

Labels let you control which version of a prompt is served in each environment without changing code.

Label	Behavior
latest	Auto-assigned to the newest version
production	Typically used for live traffic
Custom (e.g. staging, canary)	Any lowercase alphanumeric string with -, _, .

Assign labels in the Oodle UI or via the SDK. When you fetch a prompt with label="production", you always get the version currently tagged with that label — no redeployment needed to roll forward or back.

Prompt References

Prompts can reference other prompts using the @@@oodlePrompt:name=<name>|label=<label>@@@ syntax. When a prompt is fetched, Oodle resolves these references recursively and returns the fully composed content.

This lets you build modular prompt libraries — for example, a shared system instruction referenced by multiple task-specific prompts.

Managing Prompts in the UI

The Oodle UI at Agent Observability → Prompts (ap1, us1) provides a visual interface for managing prompts beyond what the SDK offers.

Prompt Types

Oodle supports two prompt types:

Type	Description
Text	A single block of text with {{variable}} placeholders
Chat	A multi-message conversation with role selection (system, user, assistant)

Choose the type when creating a new prompt. Chat prompts are useful for few-shot examples or multi-turn templates.

Folders

Organize prompts into folders for easier navigation. Create folders from the prompts list page and drag prompts between them.

Prompt Detail Page

Click any prompt to open its detail page with these tabs:

Tab	Description
Prompt	View and edit the prompt content
Config	JSON model parameters (temperature, max tokens, etc.)
Use	SDK code snippets for fetching and compiling the prompt
Metrics	Usage metrics over a configurable time range

Version History

The sidebar on the prompt detail page shows all versions. Click any version to view it, and use the diff dialog to compare two versions side by side.

Test in Playground

Click Test in Playground on any prompt to open the Playground with the prompt content pre-loaded. After iterating in the Playground, use Save as Prompt to save your changes back as a new prompt.

Best Practices

• Use labels for deploys. Fetch by label (not version number) in production code so you can update prompts without redeploying.
• Keep production stable. Test new versions with a staging or canary label before promoting.
• Use variables for dynamic content. Avoid hardcoding user inputs or context into prompt text — use {{variable}} placeholders and compile().
• Version intentionally. Each create_prompt call creates a new version. Write a commit message to track why the prompt changed.