Prompt management for teams that ship.
Write, version, test and publish your LLM prompts in one shared library, and pull them into your apps through an API. Self-hosted and MIT licensed.
git clone https://github.com/pinkroosterai/Clarive.gitcd Clarivecp .env.example .envdocker compose up -d
Code Review Assistant
Published v4Edited: the word “detailed” is replaced with “specific”.Review the following {{language}} code and provide specific feedback on bugs, performance, security, and code style:{{code_snippet}}
System message
You are an experienced senior developer performing code reviews. Analyze code for bugs, performance issues, security concerns, and adherence to best practices. Be specific and constructive.
v4
published today
v3
wording
v2
variables
- v4 published today
- v3 wording
- v2 variables
Your prompts are scattered.
Some live in the codebase, a few in Notion, the rest in Slack messages and a spreadsheet nobody has updated since October.
When a prompt change breaks production, there is no version to roll back to. When a teammate who doesn't code wants to change the wording, they file a ticket and wait for a deploy. And nobody can say whether last week's small improvement made things better or worse.
Clarive treats prompts like code — versioned, scored, edited together and served through an API — and nobody on the team has to learn Git to use it.
From first draft to the version your app calls.
Write
A Markdown editor that highlights {{template variables}} as you type, with a section for the system message and room for several prompts in one entry, run as a chain.
Keep versions
Each entry holds tabs — named variants you edit side by side — and you publish the one your app should get. Compare any two versions and restore an older one.
Read the tabs and versions docs
Refine with AI
The AI Wizard works in rounds: describe what you need, review the variations, answer its questions, and see the quality score after each round. It can pull in web search results, write the system message, or split one long prompt into a chain.
Test before you publish
The Playground runs a prompt against the models you have set up, streaming the output. Its test matrix puts tabs and models side by side with a heatmap, shows token use and estimated cost per run, and exports a PDF report.
Work as a team
Workspaces with Admin, Editor and Viewer roles, email invitations and an audit log. The editor shows who else is in a prompt, warns before you overwrite them, and helps merge conflicting edits with AI. Folders, tags, favourites and full-text search keep the library easy to find your way around.
Share a prompt
Send a published prompt as a link that needs no account, with an optional password and expiry date. Revoke the link whenever you like.
Use your own models
Pick a provider and Clarive fills in the endpoint for you: OpenAI, Anthropic, OpenRouter, Groq, Together AI, Ollama or Azure OpenAI — or point it at any other OpenAI-compatible API. The same choice appears in the setup wizard on first run.
The model browser shows what each model can do and what it costs, and with OpenRouter the prices come live from the provider.
MCP servers add tools that prompts can call in the Playground.
Clarive is for writing, testing and managing prompts. It does not trace or monitor LLM calls in production — pair it with an observability tool for that.
Your app asks for the published version.
Create an API key under Settings, API Keys, and fetch any published prompt over REST — or send values for its variables and get the rendered prompt back.
# Get a published promptcurl -H "X-Api-Key: cl_your_key_here" \ http://localhost:8080/public/v1/entries/{entryId} # Render with template variablescurl -X POST \ -H "X-Api-Key: cl_your_key_here" \ -H "Content-Type: application/json" \ -d '{"fields": {"topic": "AI safety", "tone": "professional"}}' \ http://localhost:8080/public/v1/entries/{entryId}/generateRuns on your servers, under the MIT licence.
One app container serves the web app and the API on port 8080. PostgreSQL 16 and Valkey 8 run beside it, and docker compose up -d starts all three.
The .env file needs three secrets. AI providers, email and sign-in options are set in the Super Admin dashboard after the first login, with no restart.
MIT licensed, with no open-core split and no enterprise keys holding features back.
git clone https://github.com/pinkroosterai/Clarive.gitcd Clarivecp .env.example .envdocker compose up -dFill in the three secrets in .env first — the file shows how to generate each one — then open http://localhost:8080
The first account you create becomes the super admin.