Ship better prompts
without a code push.
Version, deploy, and roll back prompts from the dashboard or via peekr deploy. A/B test variants automatically. Compliance teams stay in the loop before anything goes live.
The problem
Prompts change constantly. Your deploy pipeline doesn't.
Every prompt tweak means a code change, a PR, a review, a deploy. When a prompt regression hits production, rollback means another deploy cycle. And A/B testing two variants requires custom infrastructure nobody has time to build.
- Prompts buried in code — no history, no diff, no audit trail
- Rollback = full code deploy — slow when something breaks in prod
- A/B testing requires custom experiment infrastructure
- Non-engineers can't update prompts without a developer
Before
# prompts.py — buried in code
SYSTEM_PROMPT = """
You are a helpful assistant.
Answer using the context below.
""" # last updated: who knows
# last tested: never
# rollback: git revert + deployAfter
# agent.py — one line, always current
prompt = peekr.prompts.get(
name="rag_answer",
api_key="pk_live_…",
)
# Update from dashboard — no deploy needed
# Roll back in one click
# A/B test v3 vs v4 automaticallypeekr.yaml · GitOps for prompts
Check in your prompts. Deploy them like code.
Prefer version control over a dashboard? Define your prompts in peekr.yaml and run peekr deploy in CI. Peekr registers a new version, keeps the full diff, and activates it atomically. Rollback is peekr rollback rag_answer.
- peekr.yaml lives in your repo — full git history and PR review
- peekr deploy in CI/CD — prompts ship when code ships
- Dashboard and YAML are always in sync — no drift
- Branch-scoped staging — test new prompts on a feature branch first
# Define prompts as code
prompts:
- name: rag_answer
model: gpt-4o-mini
system: |
You are a concise assistant.
Use only the context provided.
variables: [context, language]
# Then in CI:
# $ peekr deploy --env production
# ✓ rag_answer → v7 (active)Features
Everything prompts need to run in production.
Version history
Every change is a new version. See the diff, who changed it, when, and why. Roll back to any previous version in one click.
Deploy without code
Activate a new version from the dashboard or via peekr deploy. The SDK fetches the active version at startup — no redeploy required.
A/B testing
Split traffic between two prompt variants. Peekr's observability automatically tracks cost, quality, and latency for each variant.
Variable templating
Use {{variable}} placeholders. The SDK renders them at runtime with your context, keeping templates clean and reusable across environments.
Model pinning
Pin a prompt to a specific model. When you update the prompt, the model comes along — no scattered model config across your codebase.
Non-engineer access
Product, legal, and compliance teams can update prompt wording from the dashboard. Developers review changes before activating.
SDK integration
One line to fetch. Variables render automatically.
The SDK fetches the active version at startup. Variables are rendered at call time with your context. When you activate a new version from the dashboard or runpeekr deploy, the next startup picks it up automatically — no code change needed.
import peekr
# Fetch active version — auto-updates from dashboard
prompt = peekr.prompts.get(
name="rag_answer",
api_key="pk_live_…",
)
# Render with runtime variables
system = prompt.render(
context=retrieved_docs,
language="en",
)
# Use it — model comes from the pinned version
client.chat.completions.create(
model=prompt.model,
messages=[{role: "system", content: system}]
)Prompts + Compliance
Updating a prompt means re-testing your guardrails.
Prompts and compliance are not independent concerns. When you change how an LLM responds, you may also change whether it leaks PHI, violates HIPAA safe harbor, or starts producing outputs that break your guardrail rules. Peekr ties the two together.
New prompt version activated
You update the system prompt — either from the dashboard or by running peekr deploy. A new version is registered and activated atomically.
Guardrail re-validation triggered
Peekr flags the deployment in the Guardrails panel. Your compliance rules — PHI redaction, HIPAA-safe responses, topic blocklists — run against the new prompt's output automatically.
Audit trail captured
Every prompt version links to its guardrail test results. Compliance teams see exactly which version passed which rules — and when. No spreadsheets, no manual signoff emails.
Stop hardcoding prompts.
Free tier includes unlimited prompts. No credit card required. Ship your first version in five minutes — from the dashboard or with peekr deploy.
Need observability too? See Peekr Observability →