Planck v0.1.5

custom_openai provider

New provider for any OpenAI-compatible endpoint — NVIDIA NIM, Together AI, vLLM, and others.

planck_ai

• Planck.AI.Models.CustomOpenAI — model/2 factory and runtime discovery via GET {base_url}/models
• Planck.AI.Model gains an identifier field: an uppercase tag (e.g. "NVIDIA") that derives the API key env var (NVIDIA_API_KEY) at request time — never cached
• Config.from_map/1 validates and upcases identifier; rejects values outside [A-Z][A-Z0-9]*

planck_agent

• spawn_agent provider enum extended to include "custom_openai"; base_url enforced for it the same as for ollama and llama_cpp

planck_headless / planck_cli

• SetupModal step 1 includes :custom_openai in the provider dropdown
• Step 2 gains Identifier and optional API Key fields; the API key label updates dynamically as you type the identifier
• configure_model/1 writes identifier and base_url into config; writes <IDENTIFIER>_API_KEY to .env
• Spanish translations added for all new strings

Config merge fix

JsonBinding now concatenates "models" arrays from ~/.planck/config.json and .planck/config.json instead of replacing one with the other. Previously, any local config with a models entry silently dropped
all globally configured models, causing "model not found" errors on session resume.