Amr Gharbeia
908936d4d3
- Rename gateway-provider → system-model-provider (generic :local provider, no hardcoded ollama) - Rename gateway-llm → system-model (model-request dispatcher) - Rename system-embedding-gateway → system-model-embedding - Rename gateway-manager → gateway-messaging (public api renamed to messaging-*) - Add system-model-explorer (model discovery via OpenRouter API, cached, per-slot recommendations) - Fix skill loader export: replace prefix-matching with fbound/boundp-based export (20 skills now export) - Add model-router to skill-loader exclusion list (loaded via CLI) - De-ollama: remove hardcoded assumed-available patterns from provider pipeline - Default cascade: cloud-only (openrouter, openai, groq, gemini, deepseek, nvidia, anthropic) - Env example: add LOCAL_BASE_URL, fix cascade order - All org files updated with architectural prose (literate programming)
40 lines
2.1 KiB
Org Mode
40 lines
2.1 KiB
Org Mode
#+TITLE: SKILL: Model Dispatch (org-skill-model.org)
|
|
#+AUTHOR: Agent
|
|
#+FILETAGS: :skill:model:dispatch:
|
|
#+PROPERTY: header-args:lisp :tangle ../lisp/system-model.lisp
|
|
|
|
* Architectural Intent
|
|
|
|
~system-model~ is the dispatcher. It sits between the reason pipeline and the provider backends registered in ~*probabilistic-backends*~. The reason pipeline calls ~model-request~ with a provider keyword and a model name; ~model-request~ looks up that provider's handler function, calls it, and returns the result.
|
|
|
|
This is intentionally thin. All the provider-specific logic (URL construction, API key headers, JSON parsing) lives in ~system-model-provider~. All the routing logic (which model for which task) lives in ~system-model-router~. This skill is the seam — it decouples "who to call" from "how to call them" and "why to call them."
|
|
|
|
When every provider fails (not registered, or all return errors), ~model-request~ returns an error plist with ~:status :error~. The reason pipeline's ~backend-cascade-call~ catches this and falls through to the next provider in the cascade. If no provider can serve the request, the cascade returns the "Neural Cascade Failure" signal.
|
|
|
|
~model-request~ replaces the old ~gateway-llm-request~ with the same contract but renamed for consistency with the ~system-model-*~ family.
|
|
|
|
* Implementation
|
|
|
|
** Request execution
|
|
#+begin_src lisp
|
|
(in-package :passepartout)
|
|
|
|
(defun model-request (&key prompt system-prompt (provider :openrouter) model)
|
|
"Central dispatcher for LLM requests."
|
|
(let ((backend (gethash provider *probabilistic-backends*)))
|
|
(if backend
|
|
(handler-case
|
|
(funcall backend prompt system-prompt :model model)
|
|
(error (c)
|
|
(list :status :error :message (format nil "~a Failure: ~a" provider c))))
|
|
(list :status :error :message (format nil "Provider ~a not registered" provider)))))
|
|
#+end_src
|
|
|
|
** Skill registration
|
|
#+begin_src lisp
|
|
(defskill :passepartout-system-model
|
|
:priority 100
|
|
:trigger (lambda (ctx) (getf ctx :user-input))
|
|
:deterministic (lambda (action ctx) (declare (ignore ctx)) action))
|
|
#+end_src
|