Generate fake mock data #309

Activity

1. waldekmastykarz commented on Aug 8, 2023

   

   waldekmastykarz

   on Aug 8, 2023

   Collaborator

   More actions

   That's a great idea! Recently, we released a set of mocks for all Graph endpoints based on Graph API reference docs, but I can imagine that you might need some other/more content.

   Did you give some thought how you'd see it work? Would you like the Proxy to generate the mocks on the fly (making the outcomes dynamic but less predictable) or would you rather have something like a subcommand that you can use to generate a specific set of mocks, like m365proxy mock user --numItems 100 and the command would generate hundred mocked users in a mock file that you can plug into the Proxy?
2. waldekmastykarz

   added

   enhancement

   needs peer reviewIssue needs review from other team membersIssue needs review from other team members

   on Aug 8, 2023
3. Tanddant commented on Aug 8, 2023

   

   Tanddant

   on Aug 8, 2023

   Author

   More actions

   That's a good question - my use case was actually just needing random data, I don't necessarily have an opinion on the "best way" - I guess the main thing is just mapping the Graph responses to a "mock" dataset

   👍React with 👍1Reacted by Waldek Mastykarz
4. waldekmastykarz

   added

   needs specIssue needs specificationIssue needs specification

   and removed

   needs peer reviewIssue needs review from other team membersIssue needs review from other team members

   on Dec 13, 2023
5. waldekmastykarz

   removed

   enhancement

   on Jan 16, 2026
6. waldekmastykarz commented on Feb 26, 2026

   

   waldekmastykarz

   on Feb 26, 2026

   Collaborator

   More actions

   Looked at this in the context of our current capabilities. We've come a long way since this was originally filed.

   What exists now

   • MockGeneratorPlugin: Records live API responses into mock files. This gets you real data but requires actual API access.
   • LLM integration: We have ILanguageModelClient with OpenAI and Ollama backends plus a .prompty template system. This is the natural path for generating realistic fake data.
   • Mock file format: Supports complex response bodies, headers, arrays — everything needed for rich mock data.

   Proposed approach

   Rather than integrating a data generation library like Faker.js (wrong ecosystem — we're .NET), I think we should leverage our existing LLM infrastructure:

   CLI command: devproxy mock generate --openapi <spec> --use-language-model

   This ties directly into #531. When the --use-language-model flag is set:

   1. Parse the OpenAPI spec to get paths, operations, and response schemas
   2. For each operation, send the schema to the LLM with a prompt like: "Generate a realistic JSON response matching this schema for a {operationId} endpoint. Generate {count} items for array responses."
   3. Use the response as the mock body
   4. Save to mocks file

   Count/volume support: A --count flag could control how many items to generate for collection endpoints (e.g., --count 100 for the /users scenario from the original request).

   Pagination: For large collections, we could auto-generate paginated mock responses with proper @odata.nextLink or similar pagination markers, depending on the API style.

   This approach gives the best of both worlds: realistic data without needing actual API access, and it works for any API with an OpenAPI spec.
7. waldekmastykarz commented on Oct 31, 2026

   

   waldekmastykarz

   on Oct 31, 2026

   Collaborator

   More actions

   Closing: language models now cover realistic data generation (see #531 / #2036), so a dedicated fake-data generator isn't needed.
8. waldekmastykarz

   closed this as not plannedon Oct 31, 2026