v0.64.2

Idle animation off for everyone

Highlights

• The decorative idle animation is now turned off for all users, including existing configs, via a one-time migration; re-enable it anytime with display.idle_animation = true

Improvements

• Ctrl+R reverse history search now behaves readline-style
• H1/H2 markdown headings render visually larger via underline

Fixes

• Desktop: messages typed mid-turn are queued instead of being dropped with 'already processing'

Full changelog: v0.64.1...v0.64.2

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.64.1

View clear moves to Ctrl+L

Highlights

• The view-only clear now lives on Ctrl+L (terminal-style) instead of Cmd+K, which collided with Cmd+J/K prompt navigation on macOS

Improvements

• Ctrl+L and /cls keep queued messages as well as context; only the rendered transcript is wiped
• While a diagram or diff pane is available, Ctrl+L still focuses the pane as before

Full changelog: v0.64.0...v0.64.1

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.64.0

View-only clear and cross-session prompt history

Highlights

• Cmd+K (or /cls) clears the chat view while the model keeps its full context, unlike /clear which starts a fresh session
• Ctrl+R opens a reverse search over your prompt history across all sessions

Improvements

• /cls and /clear-view are listed in /help and command suggestions, and Cmd+Shift+K stays reserved for line scrolling

Full changelog: v0.63.0...v0.64.0

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.63.0

Features

• hot-reload keybindings when config.toml changes
• disable idle animation by default
• score visual variety and neighbour contrast from the renderer
• model palette harmony as a renderer-derived graph
• add [features] auto_poke to set the auto-poke default (fixes #664)
• add [features] auto_poke to set the auto-poke default (fixes #664)
• surface colors in the config template and /config summary
• generate a whole harmonious palette from one seed color
• make every color configurable and add palette harmony scoring
• render markdown and math as typography

Fixes

• commit refresh_pinned_todos_if_needed the tick handlers already call
• gate the test-only scroll_bench::run behind cfg(test)
• repair master after a merge orphaned the skill invocation module
• isolate auth state in the mixed-catalog model picker test
• repair a header auth test that could never pass
• stop burning 60fps full frames on an animation nobody can see
• serialize gateway env tests against the shared test-env lock
• apply the corrected palette/theme pass order on master too
• don't luminance-flip the colors a user chose on light terminals
• remove the last render-path panic in harmony scoring
• make palette repair converge, and keep semantic colors meaningful
• make generated palettes colorblind-safe and hue-coherent
• land the key-to-paint metric its call site already depends on
• drop a pinned-todos override committed without its implementation
• don't let an animation frame swallow a keystroke's redraw
• restore interleave_images, dropped by the typing-backoff commit
• correct iTerm2 inline-image sequence (cell height + tmux passthrough)
• base the typing backoff on keystrokes, not provider output

Performance

• take the terminal theme query off the spawn critical path
• record key-to-paint on every paint path and expose it
• let the decorative animation step aside while the user types

Documentation

• mark why the interleave path must take the staged images
• document color configuration and harmony scoring

Other changes

• chore: rebaseline size and swallowed-error ratchets after the pinned-todos tests landed
• chore(tui): allow pinned_todos_payload_ref to be unused outside tests until its renderer lands
• test(tui): commit the pinned-todos payload tests that exercise pinned_todos_payload_ref
• Skip source auto-update quietly when local commits diverge from upstream
• release: v0.63.0
• chore: rebaseline size/panic/swallowed-error ratchets after intentional growth
• chore(tui): allow the pinned-todos state fields to be unread until their renderer lands
• test(tui): guard that every interleave send carries the staged images
• test(tui): pin the perf tier in the redraw-policy tests so they cannot flake
• style(desktop2): rustfmt render.rs and tests/delivery.rs
• desktop2: commit the frame_meter and scroll_profile modules main.rs already references
• Reframe discovery listings as editor's picks in the TUI
• Resolve run_shell merge conflict and fix stale test constructors
• test(tui): make the animated-row copy test able to fail, and restore its diagnostic
• refactor(tui-style): move the measured criteria into their own module
• test(tui-style): freeze the hand-tuned default palette
• refactor(tui): extract the idle-animation repaint policy out of run_shell
• style(tui): rustfmt the import and condition reflow CI was failing on
• test(scripts): record host load so a polluted latency measurement is identifiable
• test(tui): serialize the /colors config tests against the shared env lock
• desktop2: clamp overview navigation at edges instead of wrapping
• chore: remove legacy jcode-desktop crate and its plumbing
• desktop2: show sent versus acknowledged for user messages
• desktop2: stop remapped super+hjkl from flickering the overview
• refactor(config): move color tests into their own file for the size ratchet
• desktop2: show edits as diff cards in the transcript
• desktop2: honor reasoning display modes, with 'current' showing only the live paragraph
• desktop2: show the client version in the footnote on an empty page
• desktop2: boot-up reveal, a donut growing out of black into the window
• test(config): verify configured colors survive a real config file round trip
• desktop2: draw the session strip as nested rectangles instead of labelled bars
• test(tui-style): verify 'every color is configurable' instead of asserting it
• desktop2: remove top-right path/RAM readout from chrome row
• chore: baseline the two intentional swallowed-error patterns in palette.rs
• refactor(tui-style): split generation out of harmony, and stop panicking on the render path
• test(tui): drive /colors through the real App and dispatch table
• test(tui): prove configured colors reach a real rendered frame
• test(tui-style): calibrate harmony scoring against real terminal palettes
• ios: polish UI chrome, banners, bubbles, and code blocks
• test(tui): harnesses that separate app response from screen update
• session: attribute and de-spam bulk session-load bursts
• desktop2: align keyboard shortcuts with web/browser conventions
• desktop2: let a flick actually coast, and gate it at browser range
• test(tui): report the run loop round-trip instead of a fake arrival time
• desktop2: measure scroll feel instead of guessing at it
• test(tui): add a screen-level repro for typed characters flickering
• todo: stop a late score climb from clearing the quality gate
• todo: state the closed-feedback-loop score in one clause
• todo: trim the closed-feedback-loop description
• desktop2: make Super+HJKL actually move in the overview
• build: fail closed when the Linux compat build produces nothing or lies about its commit
• todo: rename hill-climbability to closed feedback loop
• Disable iTerm2 inline images by default (opt in with JCODE_ITERM2_IMAGES=1)
• desktop2: space the hero against the donut's worst-case ink
• desktop2: fix scrolling feel — 3 lines per wheel notch, honour the gesture phase

Full changelog: v0.62.1...v0.63.0

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.62.1

Performance

• cut the decorative animation's cost on a real session

Other changes

• release: v0.62.1
• chore(scripts): watch a release run to completion instead of assuming a pushed tag shipped

Full changelog: v0.62.0...v0.62.1

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.62.0

Features

• render markdown and math as typography

Fixes

• stop burning 60fps full frames on an animation nobody can see
• land the key-to-paint metric its call site already depends on
• drop a pinned-todos override committed without its implementation
• don't let an animation frame swallow a keystroke's redraw
• restore interleave_images, dropped by the typing-backoff commit
• correct iTerm2 inline-image sequence (cell height + tmux passthrough)
• base the typing backoff on keystrokes, not provider output
• resolve multi-word skill names in slash invocations (fixes #619) (#628)
• carry image attachments through soft interrupts (fixes #623) (#627)
• bound time-to-first-byte so a blackholed request can't hang forever (fixes #620) (#626)
• don't touch shell rc files when install dir is already on PATH (#625)

Performance

• record key-to-paint on every paint path and expose it
• let the decorative animation step aside while the user types
• stop relative cache ages from faking model-catalog changes
• stop per-frame auth lookups from blocking on credential probes
• skip duplicate remote model-catalog snapshots in the client

Other changes

• release: v0.62.0
• test(tui): harnesses that separate app response from screen update
• session: attribute and de-spam bulk session-load bursts
• desktop2: align keyboard shortcuts with web/browser conventions
• desktop2: let a flick actually coast, and gate it at browser range
• test(tui): report the run loop round-trip instead of a fake arrival time
• desktop2: measure scroll feel instead of guessing at it
• test(tui): add a screen-level repro for typed characters flickering
• todo: stop a late score climb from clearing the quality gate
• todo: state the closed-feedback-loop score in one clause
• todo: trim the closed-feedback-loop description
• desktop2: make Super+HJKL actually move in the overview
• build: fail closed when the Linux compat build produces nothing or lies about its commit
• todo: rename hill-climbability to closed feedback loop
• Disable iTerm2 inline images by default (opt in with JCODE_ITERM2_IMAGES=1)
• desktop2: space the hero against the donut's worst-case ink
• desktop2: fix scrolling feel — 3 lines per wheel notch, honour the gesture phase
• tui: make Shift+Enter actually work instead of only offering a fallback
• iOS: UX polish (graph-modeled) + production readiness (#522)
• tui: add a keymap-independent multi-line input fallback
• style: rustfmt server_events.rs after catalog-snapshot change
• test(tui): guard that detailed routes still land after a names-only catalog
• release: v0.61.2
• refactor(server): extract the catalog dedup key into its own module

Full changelog: v0.61.1...v0.62.0

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.61.1

A responsive input line while turns stream

Highlights

• Typing in the input line stays smooth while a turn is streaming: buffered keystrokes are coalesced into one frame and user input now beats server events to the loop

Improvements

• Remote-mode clients drain bursts of buffered terminal input (fast typing, key repeat, scroll wheels) into a single render, matching the local loop
• Both interactive loops poll terminal input ahead of timers and server/bus chatter, so heavy streaming can no longer starve keystrokes
• Header rebuilds no longer probe credential files on the render thread, removing periodic 40ms+ input stalls
• The desktop2 preview keeps evolving: a niri-style session overview on Super, a live RAM readout, pinned live tool cards, display-paced animations, and lower per-window memory

Fixes

• Separator spacing is preserved after LaTeX symbol commands in rendered markdown

Full changelog: v0.61.0...v0.61.1

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.61.0

Remote sessions, a destructive-command gate, and a faster desktop preview

Highlights

• /remote is a new front door for reaching a running session from another machine over the WebSocket gateway
• Destructive shell commands like recursive deletes now pause for an explicit reflection step before running, closing several bypass routes found in review
• The desktop preview gained selectable transcript text, an Alt spatial session overview, live reasoning and tool-intent streaming, and much smoother scrolling

Improvements

• Quality-gate reminders about open todos now arrive once at the end of a turn instead of nagging after every write
• Interactive model switching works across providers instead of failing when the target model lives on a different provider
• Reasoning effort options are shown for OpenAI-compatible routes that support them, and token usage is reported for their chat streams
• Failing OpenAI-compatible model-catalog refreshes back off instead of retrying hot, and each named provider keeps its own model cache
• A stall watchdog logs long silent hangs during active work so they can be diagnosed
• The slash-command registry was cleaned up: /keys no longer collides, missing aliases were restored, and help gaps were filled

Fixes

• Streaming responses no longer lose data when server-sent events split across chunk boundaries
• A turn that stops expecting a tool call but delivers none now recovers instead of stranding the session
• Pressing Enter immediately after pasting no longer sends the message prematurely
• /clear also clears the side panel
• Copilot retries transient 5xx token-exchange failures and sends the Sonnet reasoning effort correctly
• Configured context windows are honored instead of being overridden by a GPT fallback
• MCP servers started for a session now run in that session's working directory
• agentgrep respects the file field when scoping grep and find
• Session restore no longer hard-pins the Claude/OpenAI OAuth route, and a failed OpenRouter switch preserves the previous profile
• Anthropic requests that would end on an assistant turn get a continuation turn instead of erroring
• A dead terminal no longer mislabels a live session as crashed, and terminal state is restored safely on exit
• Rendered LaTeX math recovers after its image cache is evicted
• Quoted shell arguments are preserved on Windows, and the desktop build no longer breaks on Windows over Unix-only IPC

Full changelog: v0.60.0...v0.61.0

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.60.0

Discovery repair and desktop preview polish

Highlights

• Tool discovery works again for users whose config was frozen while discovery shipped opt-in, so discover_tools is registered instead of staying silently unavailable
• The desktop preview shows which provider and model is answering, and greets an empty session with an animated halftone donut and rotating composer hints

Improvements

• Discovery gained a financial-data category for market, pricing, and company-data tools
• Keybinding hints show ⌥ instead of Alt on macOS
• The desktop preview lays its text out on a consistent vertical rhythm that holds up at any window size
• A hand-written discovery opt-out is still respected, and saving your config no longer bakes today's defaults into it

Fixes

• Idle animations no longer drag the terminal UI into unnecessary full redraws

Full changelog: v0.59.0...v0.60.0

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available

v0.58.0

Max-effort reasoning streams, model context limits, and the /model picker

Highlights

• OpenAI max-effort reasoning turns no longer time out mid-thought: requests ask for reasoning summaries and the idle budget scales with reasoning effort
• The /model picker shows pretty model names and matches queries like "opus 4.8" against them
• New and unrecognized models no longer silently fall back to a 200K context window

Improvements

• Streaming idle and websocket completion budgets scale with reasoning effort (high 2x, xhigh 3x, max 4x) so long silent thinks are not mistaken for dead connections
• The client-side stall guard budgets against the largest effort-scaled server timeout, so the server's visible error always fires first
• Update failure messages collapse to one short line instead of a wall of text
• Update checks back off from GitHub API rate limits instead of retrying in a loop
• Spawned swarm workers fail loudly when they cannot use their requested model

Fixes

• Config-declared models survive a live catalog refresh
• Named profiles that point elsewhere are treated as user-declared rather than built-in
• Antigravity self-heals the missing thought_signature dead-end instead of getting stuck
• A home-directory subscribe no longer clobbers the working directory of a project session
• Kimi ids with an explicit -256k suffix resolve to 256K instead of the family default
• JCODE_TOOL_CALL_DETAILS now invalidates the cached config when it changes
• Fuzzy matching no longer rejects late matches because of the positional penalty

Full changelog: v0.57.0...v0.58.0

Platform availability

• Linux x86_64: available
• Linux aarch64: available
• macOS Apple Silicon: available
• macOS Intel: available
• Windows x86_64: available
• Windows ARM64: available
• FreeBSD x86_64: available