Skip to content
Planned interfaceSpecification; not implemented

Integrations and modules

Planned. External coding clients, creative applications, retrieval pipelines, plugin modules and companions are not operational native features in 0.0.1. Catalog source links and atlas interaction do not install or authorize them.

Supporting a client such as Codex or Claude Code requires an actual compatible server contract, authenticated endpoint, tested client version and qualified model behavior. An Anthropic-shaped Messages endpoint does not make a local model Claude. Tool-call formatting, cancellation, streaming and errors must be tested independently of a successful HTTP response.

The LM Studio integrations docs describe that product’s integrations. RegesARC uses them as a workflow reference, not a drop-in support declaration. No base URL or provider key is documented here because the RegesARC public server is not released.

Blender, Roblox, Unity, Figma, Adobe applications, browsers, filesystem tools and cloud services have different process, platform and account requirements. An integration must declare the actual upstream package/endpoint and allowed operations. A community bridge must not be labeled an official vendor product or silently assume Windows support.

Read-only design context, editing a scene, executing a script and publishing a build are distinct permissions. A preview of a setup plan should never launch an imported command or perform an external write without the applicable authorization.

The planned module system needs a versioned manifest, validated entry paths, declared capabilities, dependency/resource bounds, licensing and a lifecycle. Reject traversal and incompatible versions before activation. A plugin’s UI cannot bypass host policy; contribution hooks and removal/rollback need actual implementation.

Document retrieval needs an explicit embedding model, chunking/index versions, source citations and data-retention policy. It is not provided automatically by an MCP connection. Pets/bots are independent RegesARC designs, not ChatGPT Pets API integrations. Companion actions must use the same permission broker as any other module.

Keep proposed, reviewed, installed, enabled and connected states separate. Native implementation should begin with a small verified set and a useful failure/recovery path, then expand the ecosystem from evidence rather than broad “everything works” labels.