How-to¶

Procedures for an operator or contributor who already knows what they want to achieve: the ordered steps, the checks between them, and the way back if a step fails.

Documents¶

Document	Summary
Kudos ledger operations	Operator procedures for the kudos ledger: mode cutover, health checks, rollback, recovery, partition conversion, and event retention.
Add extended image controls to a frontend	Integrate sampler discovery, schedules, solver controls, expanded ControlNet types, and control-map annotations into an existing image frontend.
Add a sampler, scheduler, solver control, or annotator	Extend the image sampler and control-map vocabularies across hordelib, horde_sdk, AI-Horde, and the reGen bridge, in the order that keeps them compatible.
Add an image baseline	Publish the baseline record on the model reference, and add a bridge row only when a release adds the engine support.
Use the style endpoints	Read the published style contract, create text and image styles that declare a policy and placeholders, and send requests under them.
Add style authoring and styled requests to a frontend	Build style authoring controls from the published style contract, build request inputs from a style's declarations, preview with a dry run, and handle collections and style errors.
Update an existing style integration	The changes in AI Horde 5.1.13 that a client already using styles can observe, what to change for each, and how to check it.
Run the OpenAI-compatible chat front	Configure and run horde.oai, verify its model list and completions, and diagnose availability errors and lost submissions.
Send chat completions through the OpenAI-compatible front	Configure an OpenAI client for the Horde, choose a model, send chat requests and handle tools, images, reasoning, streams and errors.
Send text generation requests through the native API	Submit a prompt to the native text API, choose models and parameters, poll the status, read or cancel the generations and handle errors.
Add chat completions to a web frontend	Call the chat front from a browser, parse streamed replies, run tool calls and display model limits and errors.
Write a minimal chat worker	Run the example Python chat worker with an OpenAI-compatible backend, verify the token counter and check the worker declaration.
Serve chat jobs from a text worker	Add chat support to a text worker, declare capabilities, check prompt fit and submit replies or faults, with koboldcpp as an example.