Telegram interface ignores local model selection + can't commit files or run skills via Telegram on macOS

User nik on macOS with Ollama running locally reported three issues when using OpenHuman via the Telegram interface: Issue 1 — Telegram ignores local model selection Local Ollama model selected in the chat window works correctly in the desktop UI but is ignored by the Telegram bot, which defaults to a cloud model instead. The Telegram interface appears to have its own separate model config that doesn't inherit from the chat window selection. Workaround: manually set a telegram-specific model in config (e.g. telegram.model: "ollama/llama3.2" or OPENHUMAN_TELEGRAM_MODEL env var). Issue 2 — Cannot commit files via Telegram Agent cannot push commits or write files through the Telegram interface. Team noted this may be by design — Telegram's interface is read-only for filesystem operations. Needs confirmation of whether file commits should be supported via Telegram at all. Issue 3 — Skills not running Skills fail to execute — likely a path/config issue with SKILLS.md or the skills directory not being properly referenced. Ollama models may also struggle with structured tool calling that skills depend on. Issue 4 — Brave Search API not connecting Brave Search connector not working — team noted this requires a Brave API key set in config or env vars. Not related to Ollama being local. Resolution unconfirmed — follow-up sent 28/05/2026.