Prompt management for teams that ship.
Version control, AI refinement, quality scoring, and team workspaces for your LLM prompts.
Self-hosted. Single container. MIT licensed.
Your prompts are scattered. Some live in the codebase, a few in Notion, the rest in someone's Slack messages and a spreadsheet that hasn't been updated since October.
When a prompt change breaks production, there's no version to roll back to. When a non-technical teammate wants to tweak wording, they file a ticket and wait for a deploy. Nobody can tell you whether last week's "small improvement" actually made things better or worse.
Clarive treats prompts like code: versioned, scored, collaboratively edited, and retrievable through an API. Your whole team can use it without learning Git.
cp .env.example .env
# Fill in the 3 secrets (generation commands are in the file)
docker compose up -dOpen http://localhost:8080 and create your first account.
Pulls the pre-built image from Docker Hub. To build from source, see the development setup guide.
This isn't one-shot "generate a prompt." Clarive runs a multi-turn conversation with AI agents that keep refining until the output is genuinely good.
You describe what you need, review AI-generated variations, ask follow-up questions, and watch the quality score change with each round so you know whether your edits actually helped. Behind the scenes, Tavily web search pulls real-time context into generation, and the system can auto-generate system messages or break a monolithic prompt into a multi-step chain.
Works with any OpenAI-compatible API: OpenAI, Anthropic (via proxy), Azure, local models, whatever you've got.
Test prompts against any configured model without leaving the app. Responses stream token-by-token. Fill in {{variables}}, pick a model, tweak temperature and max tokens, then fire.
For reasoning models (o1, o3, etc.), you get the chain-of-thought alongside the answer. Multi-prompt entries show a step-by-step chain view. Run history keeps your last 20 executions for side-by-side comparison.
Every prompt moves through Draft → Published → Historical. You get the rigor of code versioning without the Git overhead.
- Word-by-word, color-coded diffs between any two versions
- One-click rollback to any previous version
- Undo/redo with snapshot history
- Concurrency protection: two people editing the same prompt won't silently overwrite each other
WYSIWYG Markdown editor built on Tiptap v3. Template variables ({{like_this}}) get highlighted inline as you type. Entries can hold multiple prompts with drag-and-drop reordering, plus a dedicated system message section.
Share any published prompt with a single URL — no account required. Optionally add a password or an expiration date. Recipients see a clean, read-only page with the prompt content and a one-click copy button. You can copy the link again later, regenerate it (invalidates the old URL), or revoke it entirely from the Manage Share Link dialog.
Roles (Admin, Editor, Viewer), multiple workspaces per account, email invitations, and a full audit log. Organize with nested folders, tags (AND/OR filtering), favorites, and full-text search.
One container. One command. One port.
nginx + the .NET API, managed by supervisor, all on port 8080. PostgreSQL runs on your infrastructure. MIT licensed with no open-core traps and no enterprise keys gating features you need.
| Feature | Clarive | Langfuse | PromptLayer | Agenta |
|---|---|---|---|---|
| Prompt versioning | Yes | Yes | Yes | Yes |
| Multi-turn AI refinement | Yes | — | — | — |
| Quality scoring with history | Yes | — | — | — |
| Web search in generation | Yes | — | — | — |
| Rich WYSIWYG editor | Yes | — | Yes | — |
| Chain decomposition | Yes | — | — | — |
| Interactive playground | Yes | Yes | Yes | Yes |
| Self-hosted (single container) | Yes | Yes (complex) | — | Yes |
| Team RBAC & audit logs | Yes | Paid | Paid | Paid |
| Public share links | Yes | — | — | — |
| Public REST API | Yes | Yes | Yes | Yes |
| MIT license | Yes | MIT (EE) | — | MIT (EE) |
| LLM observability / tracing | — | Yes | Yes | Yes |
Clarive focuses on prompt authoring, refinement, and team management. If you need LLM observability and tracing, something like Langfuse pairs well with it.
One container: nginx serves the frontend and proxies /api/ to the .NET backend. PostgreSQL and Valkey run as separate Docker services.
:8080 (nginx)
┌─────────────────────────────────┐ ┌──────────────┐
│ Clarive Container │ │ PostgreSQL │
│ ┌──────────┐ ┌────────────┐ │────▶│ 16 │
│ │ nginx │──▶│ .NET 10 API│ │ └──────────────┘
│ │ (frontend)│ │ (backend) │ │
│ └──────────┘ └────────────┘ │────▶┌──────────────┐
│ supervisor │ │ Valkey 8 │
└─────────────────────────────────┘ │ (cache) │
└──────────────┘
For the full tech stack, project structure, and design details, see docs/architecture.md.
You need Docker with Compose v2. That's it.
cp .env.example .env
# Generate and fill in the 3 secrets (commands are in the file)
docker compose up -dOpen http://localhost:8080. Frontend and API both run through that single port.
Want a specific version? Set CLARIVE_VERSION in .env (e.g., CLARIVE_VERSION=1.0.0). Default is latest.
- Create your account at the login screen. The first user becomes the super admin.
- Set up AI (optional): Go to Super Admin > AI > Providers, add your OpenAI-compatible provider with an API key, then select default models in AI > Settings.
- Invite your team: Settings > Members > Invite. Assign Admin, Editor, or Viewer roles.
- Create your first prompt: Hit "New Entry" in the library, start writing, and publish when ready.
Email is off by default. Clarive works fine without it — new users are auto-verified. If you want verification emails, password resets, and invitations delivered by email, configure it in Super Admin > Settings > Email:
- Resend — paste your Resend API key
- SMTP — point it at any SMTP server
No restart needed. Changes take effect within 30 seconds.
Add your Google OAuth client ID and secret in Super Admin > Settings > Authentication. The Google sign-in button appears automatically.
Your .env file only needs three secrets. Everything else is configured through the Super Admin dashboard after deploy.
| Variable | What | Required |
|---|---|---|
POSTGRES_PASSWORD |
Database password | Yes |
JWT_SECRET |
JWT signing key (min 32 chars) | Yes |
CONFIG_ENCRYPTION_KEY |
Encrypts secrets stored in the database | Yes |
CORS_ORIGINS |
Allowed CORS origin (your URL) | No (default: http://localhost:8080) |
CLARIVE_PORT |
Host port | No (default: 8080) |
CLARIVE_VERSION |
Docker Hub image tag | No (default: latest) |
After your first login, everything else is configurable at runtime — no restart needed:
- AI: Add providers, pick default/premium models, set a Tavily API key for web search
- Email: Choose provider (
none,resend,smtp), configure SMTP credentials or Resend API key - Google Login: Set your Google OAuth client ID and secret
- Registration: Toggle whether new users can sign up
- Rate limits: Adjust auth endpoint throttling
Full reference: docs/configuration.md.
Clarive has a public API for pulling prompts into your apps. Authenticate with an API key (X-Api-Key header), generated from Settings > API Keys in the UI.
# Get a published prompt
curl -H "X-Api-Key: cl_your_key_here" \
http://localhost:8080/public/v1/entries/{entryId}
# Render with template variables
curl -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}/generateThe first call returns the published version with its system message and prompts. The second substitutes {{topic}} and {{tone}} and returns the rendered result.
Full OpenAPI spec: docs/api-reference.yaml. Browse it interactively at /api-docs when running in development mode.
We'd love the help. The basics:
- Fork and branch
make setup && make dev— everything runs in Docker- Make changes, add tests
make test && make lint- Open a PR
Full guide with testing details, code conventions, and project structure: docs/contributing.md. Development environment setup: docs/development-setup.md.
Looking for somewhere to start? Check good first issues.
- GitHub Issues — bugs and feature requests
- GitHub Discussions — questions, ideas, general chat