Add configurable OpenAI-compatible providers (config, CLI, and settings panel)

[xmmanuellx]

Summary

Adds user-defined, OpenAI-compatible providers so Hark can talk to any Chat Completions endpoint (Ollama, LM Studio, vLLM, local gateways, or hosted proxies) — configurable via config.lua, the CLI, or the settings panel.

Fixes the missing ability to change the API base URL.

What changed

Provider definitions

• internal/config: new top-level providers table in config.lua. Each entry is { id, label, base_url }; ids must be unique, must not collide with the built-ins (openai/openrouter/xai), and base_url must be an absolute http/https URL.
• internal/ai/openai_compatible: new client that streams Chat Completions from a configurable base URL. It deliberately omits OpenRouter-specific behavior (HTTP-Referer/X-Title headers and the web search plugin).
• internal/secrets: accepts custom provider ids and resolves <ID>_API_KEY environment variables for them.

Panel + daemon management

• internal/history: migration v5 adds providers/models tables plus CRUD.
• cmd/harkd: providers_list/add/remove IPC methods; live config snapshots and an atomic reload that rebuilds the provider map without a restart. Panel-managed providers merge with config.lua providers (config wins on ID collisions).
• cmd/harkctl: provider list/add/remove commands.
• QML: a Providers section in Settings with an Add provider form (name, base URL, model, API key) and per-provider removal; custom providers now count toward hasConfiguredProvider.

Example

-- config.lua
return {
  providers = {
    { id = "local", label = "Local vLLM", base_url = "http://localhost:8000/v1" },
  },
  provider = {
    default_model = "llama-3",
    models = {
      { id = "llama-3", label = "Llama 3", provider = "local", reasoning_efforts = { "auto", "low", "medium", "high" } },
    },
  },
}

# or via the settings panel / CLI
harkctl provider add --json --id local --label "Local vLLM" --base-url http://localhost:8000/v1 --model llama-3
harkctl secret set local

Tests

• internal/config: parsing/validation of providers.
• internal/ai/openai_compatible: SSE streaming, reasoning-effort forwarding, missing-key/base-url errors.
• internal/secrets: custom-provider env fallback + name validation.
• cmd/harkd: provider add/remove handlers, collision rejection, merge/rebuild.
• QML: existing qmltestrunner suite (51 tests) still passes; qmllint clean.

go test ./... is green.

Notes / follow-ups

• A panel-managed provider currently carries one model (the one entered in the Add form). Adding multiple models per provider or editing the base URL in place is a possible follow-up; today it's remove + re-add.
• The settings panel treats a custom provider as configured if it exists, so the send action is enabled even before its API key is set (a missing key surfaces as a provider error on send).

[rotsen93]

This is what this plugin needed, i was going to ask to the main developer to do this, but i just tried, and is awesome, i can even have my local models running on this!!! Perfect