Versioned, optimized,
and deployment-ready.
Manage prompts with version history, dynamic placeholders, and auto-optimization. 8 built-in agent categories. Four research-backed optimization strategies. Full diff viewer and one-click rollback.
Never lose a good prompt
Every edit is versioned. Roll back to any prior version in one click. Fork from an old version to explore new directions without risk.
AI improves your prompts
Enable auto-optimization and let OPRO, APE, DSPy, or PromptBreeder generate better variants while you focus on strategy.
One prompt, many contexts
{{placeholders}} let the same prompt serve different companies, products, tones, and languages without duplication.
How it works
Template. Customize. Version. Optimize. Deploy.
Five stages take every prompt from idea to production. Versioning and optimization run in parallel so you iterate with confidence.
The prompt lifecycle
5 stagesTemplate
Start from 8 built-in categories
Customize
Dynamic placeholders, no hard-coding
Version
Every edit saved, every diff visible
Optimize
Let AI improve your prompts automatically
Deploy
Draft → Active → Archive lifecycle
Template
Start from 8 built-in categories
Customize
Dynamic placeholders, no hard-coding
Version
Every edit saved, every diff visible
Optimize
Let AI improve your prompts automatically
Deploy
Draft → Active → Archive lifecycle
Template
Start from 8 built-in categories
Choose from 8 pre-built agent categories — Response Composer, Stage Classifier, Action Selector, Escalation Classifier, Session Summarizer, Verifier, Retrieval, or Custom. Each template comes with a proven structure and sensible defaults.
How it works
- Response Composer, Stage Classifier, Action Selector, and 5 more
- Pre-loaded with best-practice instructions and formatting
- Custom category for fully bespoke prompts
Customize
Dynamic placeholders, no hard-coding
Inject dynamic values with {{placeholder}} syntax. Company name, user name, product, tone, language, max length — all configurable per bot or per run. Default values keep prompts working out of the box.
How it works
- 8 built-in placeholders: company_name, user_name, product_name, tone, language, and more
- Add custom placeholders for any domain-specific variable
- Default values + required/optional flags
Version
Every edit saved, every diff visible
Every save creates a version. The diff viewer highlights exactly what changed — word by word. Roll back to any prior version in one click, or fork from an old version to explore a new direction.
How it works
- Full version history with timestamps and authors
- Inline diff viewer with word-level highlighting
- One-click rollback to any prior version
Optimize
Let AI improve your prompts automatically
Enable auto-optimization and choose from 4 research-backed strategies. The system generates variants, scores them against your chosen metric, and either promotes the winner automatically or queues it for your review.
How it works
- OPRO, APE, DSPy Bootstrap, and PromptBreeder strategies
- Metrics: success_rate, verification_pass, user_satisfaction, latency
- Auto-promote or manual-review gate per prompt
Deploy
Draft → Active → Archive lifecycle
Promote prompts through a clear lifecycle. Draft while you iterate, Active when ready for production, Archive when retired. Active prompts are injected into your bot's pipeline automatically. Share across agent groups for reuse.
How it works
- Draft / Active / Archived status with clear transitions
- Auto-injected into bot pipeline when Active
- Share across agent groups for multi-bot reuse
Agent category orbit
8 categoriesPrompt Core
Response Composer
Generates the final response to the end user with tone, length, and context awareness.
Stage Classifier
Determines the current workflow stage — greeting, gathering, processing, resolution, escalation.
Action Selector
Selects the next best action to execute based on conversation context and intent.
Escalation Classifier
Decides whether to escalate to a human operator with confidence and reason.
Session Summarizer
Summarizes conversation history for memory, context windows, and handoff notes.
Verifier
Validates outputs against policies, schemas, and guardrails before delivery.
Retrieval
Guides knowledge-base search and re-ranking for grounded, factual responses.
Custom
Fully bespoke prompt for any agent or general-purpose use case you can imagine.
Optimization strategies
Four research-backed approaches to better prompts.
OPRO
Optimization by PromptingThe LLM proposes improved prompt variants from scored history, learning from what worked.
How: Meta-prompting: the optimizer reflects on past scores and generates better instructions.
APE
Automatic Prompt EngineerGenerates candidate prompts, evaluates each against a metric, and selects the best performer.
How: Search + evaluate: brute-force exploration of the prompt space with scoring.
DSPy Bootstrap
Few-shot compilationCompiles examples into optimized few-shot prompts automatically. No manual example curation.
How: Demonstrate → compile → optimize: turns examples into structured few-shot prompts.
PromptBreeder
Self-referential evolutionMutation-prompts and task-prompts co-evolve. The prompt itself learns how to improve itself.
How: Genetic algorithm: mutates, crosses, and selects prompts based on fitness scores.
Versioning
Every edit saved, every diff visible.
The version timeline shows exactly what changed, when, and by whom. Roll back to any point in one click.
Version timeline
Every edit saved. Any version restorable.
Initial template
archivedAdded tone rule
archivedTightened guardrail
archivedOptimized via APE
activeAnswers
What teams usually ask
Do I need to write prompts from scratch?
No. Each of the 8 agent categories includes a battle-tested template with pre-loaded instructions, rules, and placeholder variables. Start from the template and customize.
How does version history work?
Every save creates a new version. The diff viewer shows word-level changes. You can roll back to any prior version in one click, or fork from an old version to explore a new direction.
Can optimization run automatically?
Yes. Enable auto-optimization, pick a strategy and metric, set a frequency (hourly, daily, weekly), and the system generates, scores, and optionally auto-promotes winning variants.
What metrics can I optimize for?
Success rate, verification pass rate, user satisfaction, latency, or a custom metric you define. Each strategy scores variants against your chosen metric.
Start shipping better prompts
Build prompts that keep improving.
Create your first prompt from a template in under a minute. Add placeholders, enable optimization, and watch your bot get better with every release.