Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.abv.dev/llms.txt

Use this file to discover all available pages before exploring further.

LLM applications require constant prompt iteration. Without prompt management, each change demands code deployment, testing, and release cycles—slowing iteration and increasing risk.

​
How Prompt Management Works

ABV prompt management follows a create-version-deploy-monitor workflow:

Create and edit prompts

Create prompts through the ABV UI, SDKs (Python/JavaScript), or Public API. Prompts support variable substitution ({{variable}}), custom configuration (model parameters, tools, metadata), and two formats: text prompts (single string) and chat prompts (array of messages).

Automatic versioning

Each prompt update creates a new immutable version (1, 2, 3…). View version diffs to see exactly what changed between versions.

Label-based deployment

Labels act as pointers to specific versions. Deploy by assigning the production label. Test with staging. Roll back by reassigning production to a previous version—no code deployment required. Create custom labels for tenants, A/B tests, or any deployment strategy.

Client-side caching for zero latency

ABV SDKs cache prompts locally with automatic background refresh. Your application fetches from cache (zero network latency) while background processes check for updates.

Link prompts to traces for metrics

Connect prompts to observability traces to track which version generated each response. Analyze performance metrics (latency, token usage, costs, evaluation scores) broken down by prompt version.

​
Why Prompt Management?

Product managers and domain experts iterate on prompts directly in the ABV UI. Update prompts, test in playground, deploy to staging, and promote to production—all without engineering involvement. Reduce iteration cycles from days to minutes.
Version control with labels enables instant rollback. Assign production to a version, observe metrics, and reassign to a previous version if issues arise—rollback completes in seconds without code changes.
Use labels to maintain different versions per environment. The same prompt has staging pointing to latest version (under test) and production pointing to previous version (stable). Create tenant-specific labels for multi-tenant deployments.
Domain experts iterate on prompts in the ABV UI without writing code. Engineers review changes, test systematically, and deploy by assigning labels. Both teams work with the same prompts in the same platform.
ABV maintains complete prompt history with version diffs, timestamps, and change attribution. See exactly what changed between versions and track performance differences through linked observability data.
Create multiple labels pointing to different prompt versions. Select another label in your application, fetch the corresponding prompt, and link it to the observability trace. ABV metrics show performance by prompt version.Learn more about A/B testing →

​
Key Features

​
Core Prompt Management

Every prompt change creates an immutable new version. View version diffs, compare side-by-side, and roll back to any previous version by reassigning labels.Learn more about version control →
Labels act as pointers to specific versions. Built-in labels: production (default) and latest (most recent). Create custom labels (staging, tenant-1, variant-a) for environment separation, tenant isolation, and A/B testing. Protected labels restrict who can modify them.Learn more about version control →
Use {{variable}} syntax to insert dynamic content at runtime. Works in both text and chat prompts. Advanced features include message placeholders and composability for reusing prompt components.Learn more about message placeholders →
Store model parameters (temperature, top_p, max_tokens), tool definitions, and custom metadata alongside prompts in the config object. Change model parameters without code changes and version configuration alongside prompt changes.Learn more about prompt config →
Text prompts: Single string with variables (for completion models). Chat prompts: Array of messages with roles (for conversational models). Both support variable substitution, configuration, and versioning.Learn more about the data model →

​
Advanced Features

ABV SDKs cache prompts locally with background refresh. First request fetches from ABV; subsequent requests return cached prompts instantly. Background processes check for updates and refresh when versions change.Learn more about caching →
Test prompts interactively before deployment. Enter variables, select models, adjust parameters, and see responses in real-time. Faster iteration than deploying and testing in code.Learn more about the playground →
Test prompts systematically against datasets to measure quality before production. Compare results across prompt versions: quality scores, output differences, and failure cases.Learn more about prompt experiments →
Test multiple prompt versions simultaneously by selecting between labels and comparing metrics. ABV metrics show performance breakdown by prompt version.Learn more about A/B testing →
Configure fallback prompts embedded in your application code, used if ABV becomes unavailable and cached prompts are expired. Ensures mission-critical applications continue functioning during outages.Learn more about guaranteed availability →
Build prompts from reusable components using prompt composition. Reference other prompts as building blocks to reduce duplication and maintain consistency. Update shared instructions once to affect all referencing prompts.Learn more about composability →
Organize prompts into hierarchical folders for better management in large projects. Organize by feature, model, tenant, or environment.Learn more about folders →
Sync prompts to GitHub repositories for version control alongside code. Prompt changes create pull requests for review. Merge triggers deployment via CI/CD or manual label assignment.Learn more about GitHub integration →
Receive notifications when prompts change (new versions, label assignments, protected label modifications). Integrate with Slack for team notifications or webhooks for custom workflows.Learn more about webhooks →

​
Getting Started

Follow the quickstart to create your first prompt and use it in your application:

Get Started with Prompt Management

Step-by-step guide to creating prompts, fetching them in your application, and linking them to traces

​
Other Features

Prompts Data Model

Understand prompt structure, text vs chat formats, and configuration options

Version Control

Learn about versioning, labels, rollbacks, and deployment strategies

Link Prompts to Traces

Track metrics and performance for each prompt version through observability

Prompt Playground

Test prompts interactively before deploying to production

A/B Testing

Compare prompt variants in production with metrics-driven analysis

Caching Prompts

Understand client-side caching for zero-latency prompt fetching

Message Placeholders

Use dynamic placeholders in chat prompts for advanced templating

Prompt Experiments

Test prompts systematically against datasets before deployment