• v1.4.0 cd054af076

    v1.4.0 Stable

    olivier released this 2026-10-12 22:29:02 +02:00 | 0 commits to master since this release

    Model browser shows price and latency

    • Every model in the Browse dialog shows its price per million input and output tokens.
    • With an OpenRouter key configured, each visible model also shows time to first token (p50) and output speed (tokens/s) over the last 30 minutes, for the provider that served the most requests. The stats load once you stop typing and are cached for 10 minutes.

    Settings

    • Settings tabs are text-only and styled like Waypoint's: accent text and underline on the active tab, all tabs on one row.
    • Removed the heading that repeated the tab name at the top of each tab.

    Fixes

    • The Free badge and free-models-first sorting now work. OpenRouter sends prices as strings, so the old check never matched.
    • The model list no longer opens empty until you type.
    Downloads
  • v1.3.0 b4d962047b

    v1.3.0 Stable

    olivier released this 2026-10-12 19:32:59 +02:00 | 2 commits to master since this release

    API keys now come from Obsidian's keychain

    • Each provider in Settings → NibbleAI → API Keys now has a picker for a secret already stored in Obsidian's keychain (Settings → Keychain). data.json stores only the secret's name; the key is read when a request is made, so a rotated key takes effect immediately.
    • Upgrading: secrets that earlier versions saved as groq-api-key, openai-api-key, anthropic-api-key, openrouter-api-key and custom-api-key are linked automatically if they contain a key. Pick your secrets once otherwise.
    • NibbleAI no longer writes empty secrets to the keychain every time settings are saved. Empty entries left by older versions can be deleted in Settings → Keychain.

    Proofreading

    • A live timer runs while the model works. When it finishes, a notice shows the elapsed time, the model, the upstream host (OpenRouter), and the prompt/output/reasoning token counts. The same line is logged to the console.
    • New Reasoning (default off) and Effort (default low) settings when the provider is OpenRouter. Turning reasoning off avoids slow, token-hungry runs on models that reason by default.

    Fixes

    • A response that hits the max-tokens limit (often because the model spent its budget reasoning) now shows a clear error, instead of crashing with Cannot read properties of null (reading 'trim') or replacing your selection with a truncated answer.
    Downloads
  • v1.2.1 bf27867d20

    v1.2.1 Stable

    olivier released this 2026-09-15 02:06:58 +02:00 | 4 commits to master since this release

    Fixed

    • Check the transcription provider's API key before starting a recording, instead of only after stopping. Prevents wasting a recording when no key is configured.
    Downloads
  • v1.2.0 c6a98d4050

    v1.2.0 Stable

    olivier released this 2026-09-07 23:22:34 +02:00 | 6 commits to master since this release

    Unify AI provider selection across all features

    Every LLM-backed feature (Post-Processing, Proofreading, Title Generation, Property Filler) now offers the exact same set of AI providers, driven by one shared registry instead of five hand-duplicated provider lists.

    Added

    • Groq as a chat-completion provider. Previously Groq was transcription-only despite its OpenAI-compatible API; it's now selectable everywhere the other providers are.
    • Custom provider now works everywhere. Proofreading, Title Generation, and Property Filler previously let you pick "Custom" but had no field to set its endpoint — it silently reused Post-Processing's URL. Custom now has its own dedicated API URL/key fields, exposed on every tab that offers it, plus rows on the API Keys tab.

    Changed

    • Provider dropdowns now disable (and label) any provider whose API key isn't configured, instead of letting you pick it and fail silently at runtime.
    • Consolidated 3 duplicated provider-resolution methods plus a 4th in the transcription post-processing path into one shared helper.

    Migration

    • Existing installs are migrated automatically: the old postProcessingUrl setting becomes the new shared customApiUrl on first load if it held a genuine custom endpoint.
    • No action needed for existing single-provider setups.
    Downloads
  • v1.1.0 3abd9d9b3b

    olivier released this 2026-08-11 01:58:41 +02:00 | 8 commits to master since this release

    Downloads
  • v1.0.1 8caa86753b

    olivier released this 2026-05-25 03:46:17 +02:00 | 10 commits to master since this release

    What changed

    • Renamed "Fix punctuation and grammar with AI" to "Proofread with AI" (command + context menu)
    • Added "Show in context menu" toggle in Proofreading settings
    • API keys now named per-provider (Groq key instead of generic Transcription key)
    • Transcription provider dropdown (Groq/OpenAI/Azure/Custom) with per-provider API URL that persists independently
    • Model selection added to Title Generation and Property Filler settings
    • Configurable property presets (comma-separated list in settings)
    • Tabbed settings with Lucide icons
    • Textareas use normal font instead of monospace
    • Post-processing moved into Transcription tab
    Downloads
  • v1.0.0 af45cc762c

    olivier released this 2026-05-25 03:27:01 +02:00 | 11 commits to master since this release

    Initial release of NibbleAI — AI-powered tools for your Obsidian vault.

    Features

    • 🎙️ Transcription — Speech-to-text via Groq / OpenAI / Azure
    • ✍️ Proofreading — Fix punctuation and grammar with AI
    • 🏷️ Title Generation — Generate titles for notes
    • 📋 Property Filler — Fill frontmatter properties with configurable presets
    • 🔑 Secure API keys — Stored in system keychain via Obsidian SecretStorage
    • 🎛️ Tabbed settings — Organized settings per module
    • 🏢 Multi-provider — OpenRouter, OpenAI, Anthropic, Groq support
    Downloads