# CloudCrane

> A curation layer for products where a wrong tag hurts someone. A customer brings their records and the values those records may use, and deploys a tool their AI agent calls that cannot return a row it has not checked.

CloudCrane turns a messy catalog into data an agent can be trusted with. Rules run before any model, the model may only answer from values the customer allowed, anything it is unsure of goes to a person, and every value carries a receipt saying how it was decided. Safety fields only ever gain values, never quietly lose them, and exclusions are enforced in the database query rather than in a prompt, so a record whose value is unknown is left out rather than assumed safe. Each allowed value can carry the owner's definition, and a contract can read the owner's own PDFs, such as a supplier allergen sheet, as a source of truth, citing the page on every value it touched. A field can also read a fact, such as a weight, a date or an ID, stored only with the exact words it came from; or take its values from a value set, a long versioned list with a hierarchy, where excluding an entry also leaves out records tagged only with something broader. An answer key of values a person confirmed scores every change, and a release that would find a safety value less often than the last one is refused unless a person gives a reason. A workspace can run on its own model.

## Product

- [Playground](https://cloudcrane.ai/playground): the pipeline in five steps, on a catalog you pick, in your browser
- [Roadmap](https://cloudcrane.ai/roadmap): what runs today, what is being built, and what is named but not started
- [Docs](https://cloudcrane.ai/docs): the concepts, the pipeline end to end, and the tool an agent calls
- [Integrations](https://cloudcrane.ai/integrations): where records are read from, what can call a deployed tool over MCP or REST, the agents that sign in to a workspace (Claude, ChatGPT, Cursor, Cline), and the models a workspace can run on
- [The workspace MCP](https://cloudcrane.ai/docs/build-mcp): connect your own agent to the workspace you are building, signed in with OAuth or a build key; in Claude's connectors directory (https://claude.ai/directory/cloudcrane) and the official MCP Registry as `ai.cloudcrane/workspace`
- [Pricing](https://cloudcrane.ai/pricing): tiers by records and tool throughput, per workspace per month, in US dollars or Indian rupees, paid through Razorpay
- [Service level agreement](https://cloudcrane.ai/sla): monthly uptime targets for tool calls on paid plans, support response times, severity levels and service credits
- [Status](https://status.cloudcrane.ai): live and past uptime of the tool API, from the external monitor the SLA is measured by
- [Security and trust](https://cloudcrane.ai/trust): the controls, each one tied to shipped code

## How it works

- [AI data for safety-critical products](https://cloudcrane.ai/resources/ai-data-for-safety-critical-products): five rules, starting with unknown is never safe
- [Allergen data for AI agents](https://cloudcrane.ai/resources/allergen-data-for-ai-agents): structuring allergens a chatbot can rely on
- [Data provenance for AI agents](https://cloudcrane.ai/resources/data-provenance-for-ai-agents): what a receipt on every value holds
- [Catching an allergen that quietly disappears](https://cloudcrane.ai/resources/drift-alerts): syncs and drift alerts
- [An answer key for AI enrichment](https://cloudcrane.ai/resources/answer-key-for-ai-enrichment): golden sets and the accuracy gate
- [Bring your own model, without trusting it](https://cloudcrane.ai/resources/bring-your-own-model): what stays enforced whichever model answers
- [Test the tool, not just the data](https://cloudcrane.ai/resources/test-the-tool): scenarios that hold back a release
- [Rules before AI](https://cloudcrane.ai/resources/rules-before-ai): only ask a model what code cannot answer
- [Don't make the model your safety system](https://cloudcrane.ai/resources/safety-outside-the-model): why exclusions run in the query
- [A catalog that can't forget an allergen](https://cloudcrane.ai/resources/allergens-only-grow): ratchet fields
- [Why we took the names out of our embeddings](https://cloudcrane.ai/resources/name-free-embeddings): ranking on structure, not wording
- [Testing AI agents like software](https://cloudcrane.ai/resources/testing-voice-agents): scripted scenarios

## Proof

- [Our work](https://cloudcrane.ai/work): two shipped products, with the numbers counted from their repositories

## About

- [About](https://cloudcrane.ai/about)
- [Careers](https://cloudcrane.ai/careers)
- [Privacy](https://cloudcrane.ai/privacy)
- [Terms](https://cloudcrane.ai/terms)