Roadmap

What's coming in the next milestone. Items are listed in no particular order — priorities shift as the project evolves.

Next milestone

Personalities as .chat templates

Personalities today are Lua modules with a render() function. This works, but it means authoring a personality requires writing Lua code when really you just want to write a prompt.

The plan is to migrate personalities to .chat files that use the same {{ expression }} and {% code %} template syntax that regular .chat buffers already support. The built-in coding-assistant personality will be the first to move. Since include() already handles .chat files, the existing builder/registry indirection can be stripped away — a personality becomes a file you can read, edit, and share without touching Lua.

Slash commands

A full slash-command engine for .chat buffers. The scope includes:

• Registration — a registry for built-in and user-defined commands, with each command declaring its accepted arguments and behaviour.
• Parsing — slash commands recognised at the AST level so they participate in syntax highlighting, diagnostics, and completion.
• Preprocessing — commands fire as preprocessor rewriters before the request is sent, allowing them to modify the document (inject system prompts, toggle config, expand macros).
• Namespacing — commands are namespaced (e.g., /flemma:thinking) with automatic short-form resolution when unambiguous (just /thinking), so community plugins can define their own top-level commands without collisions.
• Config integration — commands that map to config values (like /thinking off or /thinking budget=2048) are validated and coerced through the existing config schema, giving type-checked inline configuration for free.
• Personality wiring — activating a personality via slash command (e.g., /coding-assistant language=php), with promotion from @You to @System so users don't have to manage role placement manually.
• Completion — a completion source that triggers on / at line start and enumerates available commands and their arguments.

User-overridable buffer messages

Flemma injects short text fragments into the conversation buffer — tool result placeholders, abort notices, job status text. These messages now live as flat keyed entries in gettext PO catalogues — po/flemma-harness.po for model-facing strings, po/flemma.po for user-facing UI strings — behind the flemma.messages module, and render through the templating engine.

What's still missing is the user-facing override surface: letting users merge a supplemental .po file over the base catalogue to replace individual entries by key. This would allow customising the text the model sees when a tool is denied, a request is aborted, or a background job is lost — without forking the plugin.

---

This roadmap reflects current intentions and is subject to change. Follow the changelog for what has shipped.