Version history
Every save, with a note and a name.
Save Version stores the messages and settings as they are, with your description of what changed and who saved it. Open Version History to see the full list.
Prompt management
Manage prompts like code.
Roll back in one click.
Fetch Hive is a prompt management platform for LLM apps. Write a prompt once, with variables for the parts that change. Save a version every time you edit it, deploy it as an endpoint with variants, and call it from your app, a webhook, or your AI assistant. If the model provider fails, a fallback model answers instead.
Free Developer plan · No card required · Prompt templates you can use in one click
01
ANATOMY
Messages, model, inputs, output. One LLM prompt you can test.
A prompt is a single model call that you can name, test, and reuse. You write the messages, pick the model and its settings, and decide what shape the answer comes back in. It all lives in one editor, and you can run it there before anything depends on it.
Messages and variables
01
Write system, user, and assistant messages. Put {{variables}} where the content changes, and each call fills them in from its inputs.
Model and settings
02
Pick a model from OpenAI, Anthropic, Gemini, xAI, Perplexity, and more. Set temperature or reasoning effort, max tokens, and prompt caching on Claude models.
Structured output
03
Set the response format to a JSON Schema and the answer comes back as data your code can read, not text you have to parse.
Prompt playground
04
Open Variables, fill in test values, and click Run. The editor works as a prompt playground, so you see the model response before the prompt is deployed or anything calls it.
02
VERSION
Prompt versioning: save every change, roll back the one that broke.
Prompts change often, and a small edit can change every answer. With prompt version control built in, you save a version with a note each time, see who changed what, and go back to an earlier version when the new one does worse.
Activity logs
Every call, with its tokens, cost, and time.
Each call is logged with its provider, model, status, tokens, and duration. Open one to see the request, inputs, and response, or filter by the metadata you sent with it.
03
DEPLOY
Deploy prompts as API endpoints for your app, a webhook, or your assistant.
Create an endpoint and the prompt gets a deployment name your code can call. The same deployment answers API requests, webhook triggers, and your AI assistant over MCP, and you can update what it serves without touching your code.
Endpoints and variants
Create Endpoint gives the prompt a deployment name. Call it with a variant, and Update Endpoint changes what that variant serves without a code change.
Webhook trigger
Let another service start a call with the endpoint’s webhook secret instead of your API key. Fetch Hive accepts it right away and runs the prompt in the background.
API and SDKs
Invoke a prompt with your API key from cURL or the Python, Node.js, Ruby, and PHP SDKs. Stream the response as it is generated, or wait for the final JSON.
Callbacks
Webhook calls deliver the result to your callback URL as prompt.completed or prompt.failed, signed with the same secret so you can verify it.
prompt invoke · support-summary
# call the deployed prompt from any system that can POST
$ curl https://api.fetchhive.com/v1/prompt/invoke \
-H ‘Authorization: Bearer <api key>‘ \
-d ‘{ “deployment”: “support-summary”, “variant”: “concise”,
“inputs”: { “ticket”: “Order #4821 arrived damaged” } }’
✓ req_01HZ… completed in 1.4s · response returned as JSON
Variant
concise
Model
Claude Sonnet 4.6 · fallback model ready
Usage
412 tokens · logged in Activity Logs
Next
update the variant any time, the call stays the same
WEBHOOKS AND CALLBACKS
A webhook trigger lets a service you do not control start a prompt without holding your API key. The answer comes back to your callback URL when it is ready, so nothing has to wait on the model.
Each endpoint variant has its own webhook URL
Send the webhook secret in a header, or as a query parameter
The call returns 202 with a request_id while the prompt runs
The callback is signed, so you can check it came from Fetch Hive
04
FALLBACK
When a provider fails, your fallback model answers.
Model providers have outages and rate limits. Set a fallback model on a deployment and Fetch Hive runs the same request on it when the primary model returns an error. Your app makes one call and gets one answer.
8
FALLBACK PROVIDERS
OpenAI, Anthropic, Google, xAI, Perplexity, Cohere, Moonshot, and Together AI
1
REQUEST ID
The fallback run keeps the same request, inputs, and metadata
0
CODE CHANGES
Turn it on per deployment and keep calling the same endpoint
API call
From your app
Claude Sonnet 4.6
Primary, provider error
gpt-5-nano
Fallback answers
Response
Same request ID
05
CONNECT
Edit and ship prompts from the AI assistant you already use.
Fetch Hive ships an MCP server. Connect Claude, ChatGPT, Cursor, or Codex once, and your assistant can find a prompt, change it, test it, and deploy it in plain language. Every change is saved as a version.
What your assistant can do once it is connected
Find and read any prompt in your workspace, with its messages and variables
Create a new prompt from a description in the chat
Change the model, settings, messages, or JSON schema
Every change is saved as a version you can go back to
Run a deployed prompt on test inputs, with the cost quoted first
Deploy only after you say yes. Apps keep the old version until then
WORKS WITH
Claude Code
Claude Desktop
ChatGPT
Codex
Cursor
VS Code
Gemini CLI
JetBrains
Sign in once with OAuth. Nothing is deployed without your yes.
06
START FASTER
Start from a prompt template, or save your own.
Most prompts start as a copy of one that already works. Use a template, change the wording, and it is yours. When you write one worth keeping, save it as a template for the rest of your account.
Templates · Free to use
I WANT TO
Start from something that already works.
Browse prompts built by Fetch Hive and the community. Use one in your workspace, swap the model or the wording, and run it.
Filter the marketplace to prompts only
Preview the messages before you create the prompt
Every template shows how many people use it
No account needed to browse
Save your own
I WANT TO
Turn a prompt that works into a template.
Open a prompt and choose Save as template. It is available to everyone in your account right away, and you can submit it to the Marketplace for others to use.
Shared across every workspace in your account
Start a new prompt from it with Use template
Submit to the Marketplace when it is ready to share
Free Developer plan
MOST USED PROMPT TEMPLATES
PART OF THE FETCH HIVE PLATFORM
Prompts share one workspace with your agents, workflows, and connected apps.
Use a prompt when one model call is enough. When the job needs several steps or tools, build a workflow or an agent in the same workspace, with the same models, logs, and API key.
FAQ
Questions people ask before their first prompt.
Everything else is in the prompt docs.
Prompt management is how a team stores, tests, versions, and deploys the prompts its LLM features depend on, instead of leaving them hard-coded in the app. Fetch Hive keeps each prompt in one place, like a prompt registry, with its messages, model settings, version history, and deployed variants, so you can change a prompt without shipping new code.
