v1.16.0

Highlights

• Image generation. Snotra can now create images: ask for a logo, an illustration or an icon, and it draws it with OpenAI's image model and saves it as a PNG, JPEG or WebP file in your folder. The image appears in the chat below the changed files, and a click opens it in the preview. It uses the OpenAI key you have already set up; the image model is chosen under Settings › Tool setup. OpenAI bills every image, and each one asks for your approval first (#85).
• Skills written by Snotra. Snotra can write skills into the open folder (.agents/skills/<name>/), checks each one right after writing it, and it shows up in Settings › Skills straight away. A new built-in skill explains how a good skill is put together (#160).
• Settings regrouped. "Security" is now Tools & security — what Snotra can and may do. The former Tools and MCP pages are one page, Tool setup: interpreters, keys and MCP servers (#767).
• More reliable file edits with OpenAI models. Reading a range of lines from a file failed every time with OpenAI models, and patches in a slightly different format than expected were refused. Both now work on the first try (#766, #771).
• Long tool outputs no longer overflow the context. The prompt budget now counts the system prompt and the tool definitions, and a long tool output is shortened to what still fits — the tool line in the chat says so (#169).
• The last tool steps of a run no longer go missing or stay unfinished in the chat when the machine is busy (#721).

What's Changed

• Settle a run only once its last event is in (#721) by @kkrafft1999 in #760
• Count the system prompt and tool schemas in the prompt budget, cap tool outputs on append (#169) by @kkrafft1999 in #761
• Generate images with a generate_image tool backed by OpenAI (#85) by @kkrafft1999 in #763
• Isolate the home folder in e2e runs on Windows too (#707) by @kkrafft1999 in #762
• Let Snotra write skills into the open folder (#160) by @kkrafft1999 in #765
• Add a tool-accuracy benchmark for tool-definition changes (#186) by @kkrafft1999 in #768
• Rename Security to "Tools & security" and merge Tools and MCP into "Tool setup" (#767) by @kkrafft1999 in #770
• Send OpenAI tools with strict: false (#766) by @kkrafft1999 in #769
• Read the near-unified diffs models send in apply_patch (#771) by @kkrafft1999 in #775
• Release v1.16.0 by @kkrafft1999 in #776

Full Changelog: v1.15.2...v1.16.0

v1.15.2

Highlights

• Syntax highlighting in the file preview. Code and configuration files now open coloured: JavaScript and TypeScript, JSON, YAML, Python, shell scripts, HTML and CSS, Markdown source and many more. The language is picked by the file name, and the colours follow the light and dark theme. The Source mode of Markdown, HTML and SVG files is coloured too, including code blocks inside Markdown. Selecting and copying still gives you the file exactly as it is. Plain text and logs stay plain, and a file larger than 512 KB is shown without colours so that it opens right away (#745).
• A question sent right after the app starts or after you open another folder no longer disappears into a chat that is replaced a moment later. It waits until the folder's chat is on screen and goes there (#721).
• The Markdown preview leaves more room on the left and right, growing with the width of the pane (#756).

What's Changed

• Hold a send until the chat it belongs to is on screen (#721) by @kkrafft1999 in #746
• Add a diagram of where the agent's text comes from (#747) by @kkrafft1999 in #748
• Create the safeStorage key once before the e2e files run (#689) by @kkrafft1999 in #749
• Explain the Security page flake and report what main holds (#733) by @kkrafft1999 in #753
• Report every tool event the renderer received when the tool log stalls (#721) by @kkrafft1999 in #754
• Give the Markdown preview more room on the left and right (#756) by @kkrafft1999 in #757
• Highlight source text in the file preview (#745) by @kkrafft1999 in #755
• Release v1.15.2 by @kkrafft1999 in #758

Full Changelog: v1.15.1...v1.15.2

v1.15.1

Highlights

• Choosing a reasoning level now closes the model menu, just as choosing a model does: by click, Enter or Space. The arrow keys still step through the levels and keep the menu open (#737).
• In a narrow window (below 400 px) the model and mode menus no longer close on their own while the bar under the input field rearranges its buttons (#741).
• The Markdown preview uses the full width of the pane instead of stopping at a fixed line length. Below 40 characters it stops narrowing and scrolls sideways instead (#740).
• The update dialog shows release notes formatted, with headings, lists and bold text, instead of raw Markdown. This starts with 1.15.1: the dialog that offers you this update still shows plain text, which is why these highlights go without formatting (#736).

What's Changed

• Render release notes as Markdown in the update dialog (#736) by @kkrafft1999 in #738
• Let the Markdown preview grow with the pane (#740) by @kkrafft1999 in #743
• Close the model menu after choosing a reasoning level (#737, #741) by @kkrafft1999 in #739
• Release v1.15.1 by @kkrafft1999 in #744

Full Changelog: v1.15.0...v1.15.1

v1.15.0

Highlights

• Reasoning is chosen in the chat now. Open the model menu below the input field: the Reasoning section sets how hard the model thinks, for this chat only. A model entry no longer carries a level, and the model pill shows model and level, e.g. gpt-6-luna · medium (#723).
• New chats start with medium. If your default entry used high until now, new chats start one step lower — pick high in the model menu when a question needs it. Chats from before keep the level they ran with.
• Entries that differed only in their level were merged into one; the first in your preference list stays and takes over the default and active role.
• OpenAI: GPT-5 and newer only. The add-model dialog no longer offers older models. Existing older entries keep working, without a level, and are marked "Older model, no longer offered".
• Documents as text: the agent reads PDF, DOCX, XLSX and PPTX with extract_document_text (#42).
• Wide tables in the Markdown preview fit the column instead of being cut off (#730).

Going back to 1.14.x is not supported. This version moves the model settings to a new format (v6). An older version cannot read it and starts with default settings; your model entries would have to be set up again there.

What's Changed

• Offer all seven OpenAI reasoning levels by @kkrafft1999 in #719
• Read PDF, DOCX, XLSX and PPTX text with extract_document_text (#42) by @kkrafft1999 in #717
• Wait for all four done lines again, with evidence on failure (#721) by @kkrafft1999 in #722
• Offer GPT-5 and newer only for OpenAI (#724) by @kkrafft1999 in #728
• Keep the reasoning level per chat (#725) by @kkrafft1999 in #729
• Fit wide tables in the Markdown preview to the column (#730) by @kkrafft1999 in #731
• Drop the reasoning level from model entries (#726) by @kkrafft1999 in #732
• Choose the reasoning level in the model menu (#727) by @kkrafft1999 in #734
• Release v1.15.0 by @kkrafft1999 in #735

Full Changelog: v1.14.4...v1.15.0

v1.14.4

What's Changed

• Show an arrow from the app to Applications in the DMG window by @kkrafft1999 in #710
• Make the two macOS e2e flakes report what they found by @kkrafft1999 in #711
• Move into ~/.Trash when the macOS trash refuses (#712) by @kkrafft1999 in #713
• Retry the allowed call too when the sandbox proxy is not reached (#714) by @kkrafft1999 in #715
• Release v1.14.4 by @kkrafft1999 in #716

Full Changelog: v1.14.3...v1.14.4

v1.14.3

What's Changed

• Move to undici 8 and send dispatcher requests through its own fetch by @kkrafft1999 in #701
• Bump the renderer-vendor group across 1 directory with 2 updates by @dependabot[bot] in #691
• Report a quit that does not end in the e2e helper by @kkrafft1999 in #705
• Build the macOS DMG with hdiutil instead of maker-dmg by @kkrafft1999 in #704
• Give every e2e start a home folder of its own by @kkrafft1999 in #706
• v1.14.3 by @kkrafft1999 in #708

Full Changelog: v1.14.2...v1.14.3

v1.14.3-rc.1

Full Changelog: v1.14.2...v1.14.3-rc.1

v1.14.2

What's Changed

• Keep --user-data-dir profiles out of the legacy migration by @kkrafft1999 in #690
• Move the build tool chain to Electron Forge 8 by @kkrafft1999 in #686
• Remove the temp folders of the update tests when each test ends by @kkrafft1999 in #696
• Show what the restart found when the file-changes e2e test fails by @kkrafft1999 in #698
• Bump @anthropic-ai/sandbox-runtime from 0.0.77 to 0.0.78 by @dependabot[bot] in #694
• Bump the dev-tooling group across 1 directory with 2 updates by @dependabot[bot] in #692
• Release v1.14.2 by @kkrafft1999 in #700

Full Changelog: v1.14.1...v1.14.2

v1.14.1

What's Changed

• Let a quit wait for the stores to finish writing (#679) by @kkrafft1999 in #681
• Open the middle column with the folder's README (#351) by @kkrafft1999 in #683
• Show HTML files in the preview as live, isolated pages by @kkrafft1999 in #682
• Release v1.14.1 by @kkrafft1999 in #687

Full Changelog: v1.14.0...v1.14.1

v1.14.1-rc.1

What's Changed

• Let a quit wait for the stores to finish writing (#679) by @kkrafft1999 in #681
• Open the middle column with the folder's README (#351) by @kkrafft1999 in #683

Full Changelog: v1.14.0...v1.14.1-rc.1