Centralize, version, and optimize your prompts with a full-featured prompt registry

Prompt Management

Name: Brokle
Author: Brokle

Brokle's prompt management system provides a centralized registry for storing, versioning, and deploying prompts. Manage your prompts like code with version control, A/B testing, and production deployment workflows.

Why Prompt Management?

As AI applications grow, prompt management becomes critical:

Challenge	Without Management	With Brokle
Version Control	Prompts in code, hard to track changes	Full version history with diffs
Deployment	Code deployments for prompt changes	Deploy prompts independently
Testing	Manual testing, no comparison	A/B testing, playground iteration
Collaboration	Prompts scattered across repos	Centralized registry, team access
Rollback	Redeploy previous code version	Instant rollback to any version

Core Features

Prompt Registry

Store all prompts in a central, searchable repository:

from brokle import Brokle

client = Brokle()

# Fetch a prompt by name
prompt = client.prompts.get("customer-support-v2")

# Use with OpenAI
messages = prompt.to_openai_messages({"customer_name": "Alice"})
response = openai.chat.completions.create(
    model="gpt-4o",
    messages=messages
)

Version Control

Every prompt change creates a new version:

# Get specific version
prompt_v1 = client.prompts.get("customer-support", version=1)
prompt_v2 = client.prompts.get("customer-support", version=2)

# Compare versions in dashboard
# Or get latest version (default)
latest = client.prompts.get("customer-support")

Labels (Environments)

Deploy prompts to different environments:

# Development - latest changes
dev_prompt = client.prompts.get("customer-support", label="development")

# Staging - tested changes
staging_prompt = client.prompts.get("customer-support", label="staging")

# Production - stable, reviewed changes
prod_prompt = client.prompts.get("customer-support", label="production")

Quick Start

from brokle import Brokle

client = Brokle()

# 1. Create a prompt in the dashboard or via API
prompt = client.prompts.create(
    name="greeting",
    content="Hello {{user_name}}! How can I help you today?",
    config={
        "model": "gpt-4o",
        "temperature": 0.7
    }
)

# 2. Fetch and use the prompt
prompt = client.prompts.get("greeting")
messages = prompt.to_openai_messages({"user_name": "Alice"})

# 3. Make the API call
response = openai.chat.completions.create(
    model=prompt.config.get("model", "gpt-4o"),
    messages=messages
)

import { Brokle } from 'brokle';

const client = new Brokle();

// 1. Create a prompt in the dashboard or via API
const prompt = await client.prompts.create({
  name: 'greeting',
  content: 'Hello {{user_name}}! How can I help you today?',
  config: {
    model: 'gpt-4o',
    temperature: 0.7
  }
});

// 2. Fetch and use the prompt
const fetchedPrompt = await client.prompts.get('greeting');
const messages = fetchedPrompt.toOpenAIMessages({ user_name: 'Alice' });

// 3. Make the API call
const response = await openai.chat.completions.create({
  model: fetchedPrompt.config.model || 'gpt-4o',
  messages
});

Prompt Structure

Basic Prompt

Simple single-message prompt:

name: greeting
type: text
content: |
  Hello {{user_name}}! Welcome to our service.
  How can I assist you today?

Chat Prompt

Multi-message chat format:

name: customer-support
type: chat
messages:
  - role: system
    content: |
      You are a helpful customer support agent for {{company_name}}.
      Be friendly, professional, and solution-oriented.

  - role: user
    content: "{{user_message}}"

With Config

Include model configuration:

name: code-assistant
type: chat
config:
  model: gpt-4o
  temperature: 0.2
  max_tokens: 2000
messages:
  - role: system
    content: |
      You are an expert {{language}} programmer.
      Write clean, well-documented code.

Template Variables

Use Mustache-style templates for dynamic content:

prompt = client.prompts.get("support")

# Compile with variables
messages = prompt.to_openai_messages({
    "user_name": "Alice",
    "product_name": "Brokle",
    "issue_type": "billing"
})

Variable Types

Syntax	Description	Example
`{{var}}`	Simple substitution	`Hello {{name}}`
`{{#list}}...{{/list}}`	Iteration	`{{#items}}{{name}}{{/items}}`
`{{#condition}}...{{/condition}}`	Conditional	`{{#premium}}VIP support{{/premium}}`
`{{^condition}}...{{/condition}}`	Inverted conditional	`{{^premium}}Standard support{{/premium}}`

Nested Variables

content: |
  Customer: {{customer.name}}
  Account: {{customer.account_id}}
  Plan: {{customer.plan.name}}

messages = prompt.to_openai_messages({
    "customer": {
        "name": "Alice",
        "account_id": "acc_123",
        "plan": {"name": "Enterprise"}
    }
})

Integrations

OpenAI

prompt = client.prompts.get("assistant", label="production")

response = openai.chat.completions.create(
    model=prompt.config.get("model", "gpt-4o"),
    messages=prompt.to_openai_messages(variables),
    temperature=prompt.config.get("temperature", 0.7),
    max_tokens=prompt.config.get("max_tokens")
)

Anthropic

prompt = client.prompts.get("analyst", label="production")

response = anthropic.messages.create(
    model=prompt.config.get("model", "claude-3-sonnet"),
    system=prompt.system_message,
    messages=prompt.to_anthropic_messages(variables),
    max_tokens=prompt.config.get("max_tokens", 1024)
)

LangChain

from langchain_core.prompts import ChatPromptTemplate

prompt = client.prompts.get("analyzer")
langchain_prompt = prompt.to_langchain()

chain = langchain_prompt | llm | parser
result = chain.invoke({"input": user_input})

Tracing Integration

Prompts are automatically linked to traces:

with client.start_as_current_span(name="chat") as span:
    # Fetch prompt - automatically recorded in trace
    prompt = client.prompts.get("support", label="production")

    # The trace includes:
    # - Prompt name and version
    # - Variables used
    # - Compiled messages
    response = llm.generate(prompt.to_openai_messages(vars))

View in dashboard:

Which prompt version was used
What variables were passed
Performance metrics for this prompt version

Dashboard Features

Prompt Editor

Visual editor for creating and editing prompts:

Syntax highlighting
Variable autocomplete
Real-time preview
Template validation

Version History

Track all changes:

Side-by-side diff view
Change author and timestamp
Rollback to any version
Version comments

Playground

Test prompts interactively:

Try different variables
Compare model outputs
Measure latency and cost
Save as test cases

Analytics

Monitor prompt performance:

Usage by version
Success/error rates
Latency percentiles
Cost per prompt

Workflow

Development Workflow

Deployment Labels

Label	Purpose	Who Can Edit
`development`	Active development	Developers
`staging`	Pre-production testing	Developers
`production`	Live traffic	Admins only

Production label changes require approval in enterprise plans. This prevents accidental prompt changes from affecting live users.

Best Practices

1. Use Descriptive Names

# Good
client.prompts.get("customer-support-billing-inquiry")
client.prompts.get("code-review-python")

# Bad
client.prompts.get("prompt1")
client.prompts.get("test")

2. Document Your Prompts

Add descriptions and usage notes:

client.prompts.create(
    name="customer-support",
    description="Main prompt for customer support chatbot. Uses customer context.",
    content="...",
    metadata={
        "owner": "support-team",
        "use_case": "customer_chat",
        "requires_context": True
    }
)

3. Test Before Promoting

Always test in staging before production:

# Development: rapid iteration
dev_prompt = client.prompts.get("support", label="development")

# Staging: pre-production validation
staging_prompt = client.prompts.get("support", label="staging")

# Production: stable, tested
prod_prompt = client.prompts.get("support", label="production")

4. Version Comments

Add comments when updating prompts:

client.prompts.update(
    name="support",
    content=new_content,
    comment="Improved handling of refund requests"
)

Next Steps

Playground

Test and iterate on prompts

Versioning

Manage prompt versions

Tracing

Link prompts to traces

Prompt Management

Why Prompt Management?

As AI applications grow, prompt management becomes critical:

Challenge	Without Management	With Brokle
Version Control	Prompts in code, hard to track changes	Full version history with diffs
Deployment	Code deployments for prompt changes	Deploy prompts independently
Testing	Manual testing, no comparison	A/B testing, playground iteration
Collaboration	Prompts scattered across repos	Centralized registry, team access
Rollback	Redeploy previous code version	Instant rollback to any version

Core Features

Prompt Registry

Store all prompts in a central, searchable repository:

from brokle import Brokle

client = Brokle()

# Fetch a prompt by name
prompt = client.prompts.get("customer-support-v2")

# Use with OpenAI
messages = prompt.to_openai_messages({"customer_name": "Alice"})
response = openai.chat.completions.create(
    model="gpt-4o",
    messages=messages
)

Version Control

Every prompt change creates a new version:

# Get specific version
prompt_v1 = client.prompts.get("customer-support", version=1)
prompt_v2 = client.prompts.get("customer-support", version=2)

# Compare versions in dashboard
# Or get latest version (default)
latest = client.prompts.get("customer-support")

Labels (Environments)

Deploy prompts to different environments:

# Development - latest changes
dev_prompt = client.prompts.get("customer-support", label="development")

# Staging - tested changes
staging_prompt = client.prompts.get("customer-support", label="staging")

# Production - stable, reviewed changes
prod_prompt = client.prompts.get("customer-support", label="production")

Quick Start

from brokle import Brokle

client = Brokle()

# 1. Create a prompt in the dashboard or via API
prompt = client.prompts.create(
    name="greeting",
    content="Hello {{user_name}}! How can I help you today?",
    config={
        "model": "gpt-4o",
        "temperature": 0.7
    }
)

# 2. Fetch and use the prompt
prompt = client.prompts.get("greeting")
messages = prompt.to_openai_messages({"user_name": "Alice"})

# 3. Make the API call
response = openai.chat.completions.create(
    model=prompt.config.get("model", "gpt-4o"),
    messages=messages
)

import { Brokle } from 'brokle';

const client = new Brokle();

// 1. Create a prompt in the dashboard or via API
const prompt = await client.prompts.create({
  name: 'greeting',
  content: 'Hello {{user_name}}! How can I help you today?',
  config: {
    model: 'gpt-4o',
    temperature: 0.7
  }
});

// 2. Fetch and use the prompt
const fetchedPrompt = await client.prompts.get('greeting');
const messages = fetchedPrompt.toOpenAIMessages({ user_name: 'Alice' });

// 3. Make the API call
const response = await openai.chat.completions.create({
  model: fetchedPrompt.config.model || 'gpt-4o',
  messages
});

Prompt Structure

Basic Prompt

Simple single-message prompt:

name: greeting
type: text
content: |
  Hello {{user_name}}! Welcome to our service.
  How can I assist you today?

Chat Prompt

Multi-message chat format:

name: customer-support
type: chat
messages:
  - role: system
    content: |
      You are a helpful customer support agent for {{company_name}}.
      Be friendly, professional, and solution-oriented.

  - role: user
    content: "{{user_message}}"

With Config

Include model configuration:

name: code-assistant
type: chat
config:
  model: gpt-4o
  temperature: 0.2
  max_tokens: 2000
messages:
  - role: system
    content: |
      You are an expert {{language}} programmer.
      Write clean, well-documented code.

Template Variables

Use Mustache-style templates for dynamic content:

prompt = client.prompts.get("support")

# Compile with variables
messages = prompt.to_openai_messages({
    "user_name": "Alice",
    "product_name": "Brokle",
    "issue_type": "billing"
})

Variable Types

Syntax	Description	Example
`{{var}}`	Simple substitution	`Hello {{name}}`
`{{#list}}...{{/list}}`	Iteration	`{{#items}}{{name}}{{/items}}`
`{{#condition}}...{{/condition}}`	Conditional	`{{#premium}}VIP support{{/premium}}`
`{{^condition}}...{{/condition}}`	Inverted conditional	`{{^premium}}Standard support{{/premium}}`

Nested Variables

content: |
  Customer: {{customer.name}}
  Account: {{customer.account_id}}
  Plan: {{customer.plan.name}}

messages = prompt.to_openai_messages({
    "customer": {
        "name": "Alice",
        "account_id": "acc_123",
        "plan": {"name": "Enterprise"}
    }
})

Integrations

OpenAI

prompt = client.prompts.get("assistant", label="production")

response = openai.chat.completions.create(
    model=prompt.config.get("model", "gpt-4o"),
    messages=prompt.to_openai_messages(variables),
    temperature=prompt.config.get("temperature", 0.7),
    max_tokens=prompt.config.get("max_tokens")
)

Anthropic

prompt = client.prompts.get("analyst", label="production")

response = anthropic.messages.create(
    model=prompt.config.get("model", "claude-3-sonnet"),
    system=prompt.system_message,
    messages=prompt.to_anthropic_messages(variables),
    max_tokens=prompt.config.get("max_tokens", 1024)
)

LangChain

from langchain_core.prompts import ChatPromptTemplate

prompt = client.prompts.get("analyzer")
langchain_prompt = prompt.to_langchain()

chain = langchain_prompt | llm | parser
result = chain.invoke({"input": user_input})

Tracing Integration

Prompts are automatically linked to traces:

with client.start_as_current_span(name="chat") as span:
    # Fetch prompt - automatically recorded in trace
    prompt = client.prompts.get("support", label="production")

    # The trace includes:
    # - Prompt name and version
    # - Variables used
    # - Compiled messages
    response = llm.generate(prompt.to_openai_messages(vars))

View in dashboard:

Which prompt version was used
What variables were passed
Performance metrics for this prompt version

Dashboard Features

Prompt Editor

Visual editor for creating and editing prompts:

Syntax highlighting
Variable autocomplete
Real-time preview
Template validation

Version History

Track all changes:

Side-by-side diff view
Change author and timestamp
Rollback to any version
Version comments

Playground

Test prompts interactively:

Try different variables
Compare model outputs
Measure latency and cost
Save as test cases

Analytics

Monitor prompt performance:

Usage by version
Success/error rates
Latency percentiles
Cost per prompt

Workflow

Development Workflow

Deployment Labels

Label	Purpose	Who Can Edit
`development`	Active development	Developers
`staging`	Pre-production testing	Developers
`production`	Live traffic	Admins only

Production label changes require approval in enterprise plans. This prevents accidental prompt changes from affecting live users.

Best Practices

1. Use Descriptive Names

# Good
client.prompts.get("customer-support-billing-inquiry")
client.prompts.get("code-review-python")

# Bad
client.prompts.get("prompt1")
client.prompts.get("test")

2. Document Your Prompts

Add descriptions and usage notes:

client.prompts.create(
    name="customer-support",
    description="Main prompt for customer support chatbot. Uses customer context.",
    content="...",
    metadata={
        "owner": "support-team",
        "use_case": "customer_chat",
        "requires_context": True
    }
)

3. Test Before Promoting

Always test in staging before production:

# Development: rapid iteration
dev_prompt = client.prompts.get("support", label="development")

# Staging: pre-production validation
staging_prompt = client.prompts.get("support", label="staging")

# Production: stable, tested
prod_prompt = client.prompts.get("support", label="production")

4. Version Comments

Add comments when updating prompts:

client.prompts.update(
    name="support",
    content=new_content,
    comment="Improved handling of refund requests"
)

Next Steps

Playground

Test and iterate on prompts

Versioning

Manage prompt versions

Tracing

Link prompts to traces

Prompt Management

Playground

Versioning

Tracing

On this page

Prompt Management

Playground

Versioning

Tracing

On this page