Sherlock lets you try AI models in a web interface, build assistants that use your documents, and send model requests from application code. Choose the web interface for interactive work or the API for scripts and integrations. Both use a Sherlock project.
Start with the introduction, practical uses, and terminology, then prepare account access through My Portal and select a project. These steps are shared by the UI and API workflows.
Use Test Playground → Chat for a first conversation or OCR to extract text from a file. To answer questions about your own documents, create a knowledge base, connect a search agent, and configure an assistant. Test the saved assistant in the Playground, then inspect consumption and project events.
Create an API key in the console, then use the getting-started guide to install the client, configure authentication, and send a first request. The model and endpoint references cover model discovery, conversation history, embeddings, and image inputs. The final guide is for updating an existing Assistant module through its separate configuration API.
These worked examples use Sherlock with other services. Choose one to compare model quality, translate queued text with a Python worker, or process translations with Functions and Object Storage. Each article includes its own prerequisites and code; Sherlock access and credentials for the other services are prepared separately.
Compare models and prompts with AI Experiments — classify the same support messages with two models and two prompts. Record accuracy and response times in MLflow to compare the results.
Translate queued text with a Python worker — send jobs through Message Broker and process them with a separate Python program. Start with local text processing, then call Sherlock for translation and save each result as a JSON file on the worker’s computer.
Translate dataset descriptions with Functions and Object Storage — test one translation over HTTP, then use a RabbitMQ trigger to process one job per target language. The function reads the source from Object Storage and saves completed translations there.
