I wanted to explore creating a human-like poker experience with LLM-powered opponents that tell themselves a story about each move, allowing them to bluff and be bluffed. It became an interesting challenge evaluating models and prompting them to behave as entertaining opponents in a cost-effective and responsive manner.
One can clone the project, fetch the submodule with original C repo and run "cargo test" to test the library with all the official tests from the C library. The fuzz directory has instructions for fuzzing
Generate Model Context Protocol (MCP) servers from OpenAPI / Swagger specifications to let AI assistants like Claude and Cursor use the APIs in a structured way that you control. Servers are generated to a single easily-inspectable TypeScript file - locally in your browser.
Flushout isn't tied to any particular cloud provider (my backend is running in AWS), and lets you preserve a history of every update to the model to make it easy to implement undo or just display historical updates to the document. As Flushout can also send individual updates to sync models, it may also be more efficient than sending snapshot updates (especially for large documents), but don't know how Firestore transfers its model updates internally...
Can multiple users share a bulk post? Curious if it would be possible to set up a site that reduces latency by grouping post actions from multiple users, possibly even letting them post for free by displaying ads in the front-end.