8000 Releases · kiran-brahma/writing-tool · GitHub
[go: up one dir, main page]

Skip to content

Releases

Obelus v0.2.0

Choose a tag to compare

@kiran-brahma kiran-brahma released this 27 Sep 04:13

Added

  • Reasoning effort now offers Off (no thinking), Minimal, Extra high and Max alongside Low, Medium
    and High, with clearer labels. It now reaches Anthropic and Gemini Connections too, not only the
    OpenAI-compatible ones.

Fixed

  • Thinking models no longer run out of output budget before answering a model pass. Passes stop
    asking the model for character offsets, which it spent its whole budget counting; Obelus finds
    each quote itself. The Topic strings pass also caps its findings at ten and asks for short quotes.
    An existing library keeps its saved Starter passes, so use Restore the Starter pack to pick up
    the new prompts.
  • Anthropic Connections work with current Claude models. Obelus no longer sends a temperature,
    which Claude models released after Opus 4.6 reject.

Obelus v0.1.0

Choose a tag to compare

@kiran-brahma kiran-brahma released this 26 Sep 13:22

The first tagged release. It covers the v1 spec through v1.3 (the page and the mark), as deployed to
Cloudflare Workers.

Added

  • Rich-text editor with a library of documents, tags, search and status; automatic and milestone
    revisions; a word-level diff; Markdown import and export; whole-library backup and restore; and a
    single-document bundle.
  • Rule passes that run on save, offline and with no API key: hedges and intensifiers,
    nominalizations, throat-clearing openers, wordy constructions, repeated words and openers, the
    passive pass at note severity, and sentence-rhythm metrics. Their word lists and patterns are
    editable, and the Voice list silences rules.
  • Model passes that return findings anchored to quotes, with praise struck through rather than
    hidden. They include per-pass screening frames, the reader pass, and a pass workbench whose
    assistant works on prompts and never on prose.
  • The Judge, which compares two revisions twice with the labels swapped, defaults to a different
    Connection from the Critic, and records a session-only calibration prediction.
  • Connections for OpenAI, Anthropic, Google Gemini, OpenRouter, local Ollama and Custom endpoints,
    with output ceiling and reasoning effort controls, cost estimates, and bounded calls so a stalled
    Provider cannot hold the mutation lock.
  • The Working order rail with Findings and Judge modes, the Current Finding, a Callout on each
    Highlight, margin marks beside Paragraphs with open Findings, and the Outline in the left margin
    on wide screens.
  • Literata type, a reading measure, the Status line, a dark scheme with a Light | Dark | System
    choice, a narrow layout where the Rail overlays the prose, plain-language glosses, and
    accessibility work on the rail, announcements and contrast.
  • A Privacy page that lists where data lives and how to verify the single-origin claim in DevTools.
  • An offline shell: the app opens with existing documents without a network.
  • PNG app icons (192, 512, maskable 512) and an Apple touch icon, so the app installs with its own
    icon on iOS, Android and desktop Chrome.
  • README instructions for installing Obelus as an app, and for allowing the deployed origins in
    OLLAMA_ORIGINS so local Ollama, and Ollama Cloud models through the local daemon, work from the
    deployed site.

Changed

  • The web app manifest has a stable id, and its theme_color matches the light background, so an
    installed window's title bar no longer flashes dark.
0