github can1357/oh-my-pi v18.0.5
on GitHub

3 hours ago

@oh-my-pi/pi-agent-core

Fixed

  • Corrected remote compaction summaries so they accurately report the number of input tokens processed.

@oh-my-pi/pi-ai

Breaking Changes

  • Renamed the exported stream-retry helper from withEmptyCompletionRetry to withReplaySafeStreamRetry and added retry policy options for empty completions and provider errors. Consumers using the old helper must migrate.

Added

  • Added browser-based Sign in with OpenRouter using OAuth PKCE, while retaining support for pasted OpenRouter API keys and redirect URLs for remote sessions.
  • Added /login API-key authentication for DeepInfra and Yolo-Auto, including validation against each provider before the credentials are accepted.

Fixed

  • Fixed DeepSeek vision models from losing image input while keeping image parts stripped for text-only DeepSeek endpoints.
  • Fixed OpenAI-compatible gateways that report uppercase completion reasons such as STOP or MAX_TOKENS; these are now classified correctly, including mapping MAX_TOKENS to a length limit.
  • Fixed provider message-count limit errors being treated as unrecoverable payload errors instead of recoverable context overflows.
  • Improved Codex WebSocket continuations so rate limits, throttling, and compatible mode changes preserve valid response continuations instead of unnecessarily replaying the full context.
  • Fixed Codex WebSocket cleanup failures caused by already-closed sockets.
  • Added safe retries for transient mid-stream socket closures across OpenAI Responses, Chat Completions, Azure OpenAI Responses, and Codex SSE when no replay-unsafe output has been emitted.
  • Fixed usage and cost reporting for OpenAI-compatible gateways backed by Vertex AI or Gemini by recognizing cached prompt tokens reported through cachedContentTokenCount.

@oh-my-pi/pi-catalog

Added

  • Added built-in DeepInfra provider support (deepinfra, DEEPINFRA_API_KEY) with live model discovery, including chat models, context windows, pricing, cache-read pricing, vision input, and reasoning-effort capabilities.
  • Added built-in Yolo-Auto provider support with the flat-rate deepseek-flash-v4 model and live model discovery.

Fixed

  • Fixed the Synthetic provider’s default model to use hf:zai-org/GLM-5.2 instead of the retired hf:zai-org/GLM-5.1.

@oh-my-pi/pi-coding-agent

Added

  • Added append-only transcript declarations and stable-row APIs for components with immutable history prefixes.
  • Added the :img read selector to rasterize local SVG and SVGZ files for vision input.
  • Added side-by-side image and SVG previews to omp git, including Git LFS object resolution and clear placeholders for unavailable or unsupported binary content.
  • Added the omp if-bench command for zero-tool instruction-following and working-memory benchmarking across models, with live progress and ranked results.
  • Added q to quit the git TUI.
  • Added advanced whitespace filtering to the git TUI, including formatting-only changes and import-only changes in TypeScript, JavaScript, Rust, and Go.
  • Improved the git TUI sidebar by compressing single-child directory chains and separating new or untracked files from tracked changes.
  • Added Yolo-Auto to /login and documented the YOLO_AUTO_API_KEY environment variable.
  • Updated the OpenRouter /login flow to support browser-based sign-in and automatic API-key provisioning, while retaining support for pasted sk-or-… keys.
  • Added DeepInfra support for the image_gen and tts tools, including provider selection and MP3 or WAV output for text-to-speech.

Changed

  • Standardized completed edit results with hashline-style paths and numbered previews across edit modes.
  • Improved omp git responsiveness with immediate file rendering, progressive syntax highlighting, and deferred large-commit statistics.
  • Documented that retry.maxDelayMs: 0 permits provider-requested quota waits to continue until automatic retry, rather than enforcing a wait ceiling.
  • Expanded git TUI navigation and file-management shortcuts, including refresh, stage/unstage, directory operations, hunk and file navigation, pane movement, diff-view selection, commit-form access, and paging.

Fixed

  • Fixed race condition where tunnel startup was incorrectly reported as failure on quick process exit
  • Fixed Obsidian theme task instructions and usage-limit text becoming unreadable against dark backgrounds.
  • Fixed marketplace-installed plugins failing to discover their rules/ directories.
  • Fixed advisor notes in /tree displaying internal XML wrappers instead of readable text.
  • Fixed successful agent and subagent results being discarded when cleanup exceeded its deadline.
  • Fixed exiting plan mode mid-turn so the active turn now stops immediately.
  • Fixed Windows workstation context reporting a virtual display adapter instead of the physical GPU.
  • Fixed numbered selector menus such as /review ignoring digit-key selection.
  • Fixed the welcome screen failing to reflow after terminal resizing.
  • Fixed transcript layout issues that could clip assistant text, leave stale tool cards, disrupt scrolling, or make large-session rendering and shutdown unreliable.
  • Fixed streamed assistant responses failing when later provider updates revised earlier Markdown.
  • Fixed cache-miss status placement after streamed assistant output.
  • Fixed /model role-cycle icons overlapping their ordinal on full-width terminals.
  • Fixed constrained /collab QR codes rendering as empty rows; the browser URL hint is now shown instead.
  • Fixed hub list and child peer rosters incorrectly including parked agents in model context and restored accurate status counts.
  • Fixed browser relay clients hanging when enabling the Runtime domain on a shared tab.
  • Fixed interactive TTSR interruptions being displayed as errors after successful rule injection.
  • Fixed fast tool completions leaving a persistent running summary that obstructed later output.
  • Fixed the Windows console welcome header being duplicated after cold launch.
  • Fixed git TUI hunk navigation feedback when the sidebar has focus and preserved file selection after staging or unstaging.
  • Fixed long sessions becoming unrecoverable when providers reject histories exceeding message-count limits.
  • Improved /dump output with readable system-notice titles and XML-fenced raw payloads.
  • Fixed kernel sessions failing to recover when cancellation was reported by a dead kernel.
  • Applied advisor tool-call loop limits to prevent repeated failing calls from continuing without bound.
  • Fixed lsp rename_file incorrectly reporting inaccessible paths as nonexistent and mishandling uncertain destination checks.
  • Fixed LSP clients with different launch or initialization configurations incorrectly sharing one process; reloading now replaces superseded clients.
  • Fixed the browser relay leaving Chrome's debugging infobar attached after the last client released a tab.
  • Fixed auto-retry countdowns appearing frozen during long provider-requested waits.
  • Fixed the Todo HUD becoming stale after viewing a subagent and returning to the main session.
  • Fixed child-task artifact links and duplicate hub jobs result bodies.
  • Fixed omp update showing a POSIX reinstall command on Windows after a package-rename migration failure.
  • Preserved thinking.requiresEffort: false in custom model configuration so supported local Qwen templates can disable thinking explicitly.
  • Fixed project settings from shared capability files being able to replace an entire settings group when a conflicting non-object value is present; the conflicting value is now ignored with a warning.
  • Jina Reader requests now use configured credentials for higher authenticated rate limits while remaining available anonymously.
  • Fixed advisor session persistence and loading performance for repeated retries and unusually large advisor transcripts.
  • Fixed failed browser.open calls leaving OMP-spawned application processes running when no tab could be acquired.
  • Improved browser-handle failures with prompt, operation-specific timeout errors instead of waiting for the entire browser cell.
  • Fixed autonomous runs becoming idle after thinking-only length stops during speculative handoff.
  • Fixed completed assistant replies disappearing from the live transcript under viewport pressure.
  • Accelerated SHA-2 and SHA-3 checksums on supported ARM64 hardware.
  • Fixed large MCP tool payloads being stored redundantly on disk.

@oh-my-pi/pi-natives

Added

  • Added asynchronous, size-bounded SVG-to-PNG rasterization for terminal media previews.
  • Added the DiffStream API for processing text and byte input incrementally, opening files asynchronously, reporting stable-prefix progress, generating exact unified diffs, and warming syntax grammars asynchronously.

@oh-my-pi/pi-tui

Breaking Changes

  • Renamed the public TerminalFrameProvider.resetHistory method to beginHistoryReplay.

Added

  • Loader messages can now be provided as a function, allowing dynamic labels such as live countdowns to update on each spinner tick while preserving the existing behavior for static strings.

Changed

  • Improved history replay and terminal output handling so replayed content is rendered efficiently and complete replay results are written together.

Fixed

  • Fixed graceful shutdown so finalized output is correctly retired before handing control back to the shell.
  • Fixed terminal scrollback corruption during shutdown, tmux pane zoom and resize, and destructive screen resets, preventing duplicated frames, lost history, and stale transcript re-streaming.
  • Fixed streaming Markdown rendering at chunk boundaries to preserve CommonMark emphasis behavior for Unicode text and correctly recognize GFM tables as they are completed.

@oh-my-pi/pi-utils

Added

  • Added stableStringifyJson for deterministic serialization of nested JSON-shaped data.

Fixed

  • Fixed managed Chrome-for-Testing installation failures when extracting the trusted browser download.

What's Changed

  • feat(catalog,ai,coding-agent): add DeepInfra provider by @ats3v in #9223
  • fix(catalog): point the Synthetic default at a model the catalog still has by @Robinbinu in #9611
  • fix(ai): recognize provider message history limits by @roboomp in #9630
  • feat(tui): tick the auto-retry countdown down instead of freezing by @sgfvamll in #9634
  • fix(ai): prevent Codex WebSocket close crash by @ephraimduncan in #9641
  • fix(task): prevent unreadable and duplicate results by @roboomp in #9649
  • fix(ai): preserve codex chains across service tiers by @alphastorm in #9650
  • fix(collab): hide QR when transcript clips below full height by @Giardi77 in #9658
  • fix(ai): exempt DeepSeek vision SKUs from text-only image guard by @chuzui in #9436
  • fix(coding-agent): authenticate Jina Reader requests by @poorpaper in #9468
  • fix(agent): resumed maintenance after speculative handoff by @roboomp in #9530
  • fix(settings): non-object project leaves no longer shadow settings groups by @sjawhar in #9532
  • fix(browsers): extract managed chrome past 64 MiB member cap by @roboomp in #9538
  • fix(browser): guarded handle actions fail fast instead of stalling the cell by @roboomp in #9539
  • fix(browser): reaped spawned apps after failed opens by @roboomp in #9540
  • test(coding-agent): wait for complete detached PID writes by @pedropaulovc in #9542
  • fix(builtins): enabled arm hash acceleration by @roboomp in #9556
  • fix(advisor): dedupe replayed transcript persistence and unblock resume by @roboomp in #9557
  • test(coding-agent): remove load-sensitivity from three native/unit suites by @pedropaulovc in #9561
  • fix(ai): treat uppercase OpenAI finish_reason as case-insensitive by @yiliang114 in #9566
  • fix(coding-agent): refresh todo HUD when re-attaching a session by @roboomp in #9575
  • fix(tui): keep final replies visible under viewport pressure by @roboomp in #9577
  • fix(models-config): preserve thinking.requiresEffort from models.yml by @serkud in #9578
  • fix(compaction): describe remote compaction summary as processed input tokens by @roboomp in #9587
  • fix(coding-agent): suppress TTSR interruption errors by @roboomp in #9589
  • fix(tui): separated model role cycle ordinal by @roboomp in #9592
  • perf(tui): skip tail re-lex on inert markdown deltas (fast-tail splice) by @szavadsky in #9598
  • fix(coding-agent): stop cold launch clearing native scrollback twice by @roboomp in #9600
  • test(utils): skip resolveWindowsShell cmd.exe fallback under WSL by @roboomp in #9606
  • test(coding-agent): stabilize streaming edit precheck timing by @roboomp in #9607
  • fix(ai): preserve Codex append state after throttling by @serverinspector in #9615
  • fix(update): use platform-aware reinstall hint in rename migration by @roboomp in #9621
  • docs(coding-agent): document retry.maxDelayMs 0 disabling the wait cap by @sgfvamll in #9632
  • fix(ai): retry replay-safe OpenAI stream failures by @roboomp in #7981
  • fix(browser): replay Runtime contexts per relay session by @roboomp in #9617
  • feat(catalog,ai,coding-agent): add Yolo-Auto provider with /login flow by @H4vC in #9665
  • fix(task): preserve successful subagent yield across deferred cleanup by @roboomp in #9673
  • fix(prompting): prefer physical Windows GPUs by @roboomp in #9676
  • fix(coding-agent): drop duplicated rawContent from spilled tool results by @roboomp in #9689
  • fix(coding-agent): interrupt in-flight turn on plan-mode exit by @roboomp in #9703
  • fix(discovery): load marketplace plugin rules by @roboomp in #9704
  • fix(tui): hide advisor XML in session tree by @roboomp in #9709
  • fix(tui): select numbered menu option by pressing its digit by @roboomp in #9711
  • fix(ai): read vertex cachedContentTokenCount as cache-read source by @roboomp in #9714
  • fix(theme): improved obsidian muted text contrast by @roboomp in #9715
  • fix(tui): prioritize transcript text over tool cards under viewport pressure by @roboomp in #9723

New Contributors

Full Changelog: v18.0.4...v18.0.5

Check out latest releases or
releases around can1357/oh-my-pi v18.0.5

Don't miss a new oh-my-pi release

NewReleases is sending notifications on new releases.

Get notifications