Skip to main content

Changelog

All notable changes to Ante, newest first. Every version is published as a GitHub release, and you can install any of them with ante update --version <V>.

v0.preview.65​

LatestGitHub release

Added​

  • Add a /pty window that attaches a native terminal side panel
  • Render LaTeX math in the TUI via a delimiter normalizer and TeX→Unicode converter
  • Add an auto compact setting

Changed​

  • --tools now sets the base tool set and --include-tools adds on top of it
  • Serialize concurrent Edit/Write calls that target the same file instead of only warning
  • Speed up Grep by dropping per-file parallel fan-out (~2× faster)
  • Use catalog keys as provider IDs
  • Cap skill catalog descriptions rendered into the system prompt
  • Extend short-prompt coverage to the system prompt and more tools
  • Harden release publication identity and atomicity

v0.preview.64​

GitHub release

Added​

  • Add an Antix auth login command

Changed​

  • Wrap tool header lines at the viewport edge instead of clipping them
  • Dependency updates

Fixed​

  • Detect when iTerm2 blocks scrollback clearing: purge by default when verified safe, show a notice when blocked
  • Exit the TUI cleanly on broken stdout pipes
  • Omit absent optional arguments from tool call displays

v0.preview.63​

GitHub release

Added​

  • Add Tencent Hy3 to the OpenRouter model catalog and pricing table
  • Include step count in turn end events

Changed​

  • Markdown render improved by 93% for long streams
  • Remove Qwen 3.7 Max/Plus from the OpenRouter catalog (endpoints no longer accessible); they remain available via Antix
  • Render markdown lists directly instead of via tui-markdown
  • Inset all TUI rows two columns from the window edge
  • Consolidate the ante-guide into a bundled, model-only skill
  • Warn when concurrent Edit/Write mutations target the same file path
  • Prune older built-in model catalog entries

Fixed​

  • Reject tool calls with malformed arguments before execution and return the parse error so the model can retry
  • Handle tool calls truncated at the output-token limit gracefully across all providers, continuing the turn instead of surfacing a hard error
  • Drop malformed ambient replies instead of leaking model reasoning

v0.preview.61​

GitHub release

Added​

  • Client-side PDF reads: page rendering, text extraction, and page bounding
  • CLI option to load a system prompt from a file
  • WebFetch spills oversized responses to a cache file instead of truncating

Changed​

  • Corrective tool errors and no line clipping in Read
  • Lenient tool path resolution and normalized permission matching
  • TUI rendering performance and composer text contrast improvements

Fixed​

  • Fall back to a usable provider instead of localhost for unknown provider ids
  • Bound PDF render dimensions against hostile input

v0.preview.60​

GitHub release

  • Reconnect and resume when an LLM stream dies mid-response — even before any answer text has arrived — instead of ending the turn with an error
  • Fail streams that end without a proper stop signal as retryable instead of hanging the turn
  • Honor server Retry-After hints, rebalance retry budgets toward the visible reconnect layer, and log every retry decision
  • Add Kimi K3 support to Antix and the OpenRouter model catalog
  • Add Meta Muse Spark 1.1 to the OpenRouter model catalog
  • Add a short_prompt toggle to the /config dialog
  • Show effort next to the model in the default status line order
  • Scroll the status line picker on short terminals instead of clipping items, and hint that Claude Code-compatible status_line_command scripts are supported
  • Exit cleanly when stdout closes mid-output (e.g. when piping into head)

v0.preview.59​

GitHub release

  • Add a --no-session-save CLI flag that skips session event logs and resumable snapshots for TUI and headless sessions, and propagates to subagents
  • Add a --short-prompt option (CLI flag and settings.json) that uses compact tool descriptions to shrink the system prompt
  • Add xAI Grok 4.5 support to Antix
  • Support GPT-5.6 on OpenAI subscriptions
  • Dependency updates

v0.preview.58​

GitHub release

  • Add a /config settings dialog to the TUI
  • Fall back to CLAUDE.md for project instructions when no AGENTS.md exists
  • Fix persisted model effort restoration and ambient settings drift
  • Preserve parse error details when loading user model catalogs
  • Render agent output text in the theme foreground color
  • Improve session shutdown reliability by cancelling and draining background session tasks before the session ends
  • Unify TUI state ownership across modals, pickers, the status line, and terminal size to prevent stale rendering after resizes and modal transitions
  • Update the pinned offline llama.cpp engine to build b9986
  • Remove a stale xAI model catalog alias
  • Refresh the mascot logo pixel art

v0.preview.57​

GitHub release

  • Persist Shift+Tab-selected permission modes as the default for new sessions while keeping Yolo session-only
  • Add curated contextual tips beneath the working spinner, with cooldowns and an opt-out setting
  • Fix turns that could hang after the model completed by stopping stream collection at MessageStop
  • Improve LLM error recovery with structured error kinds, actionable /connect guidance for expired OAuth, bare 403 classification, and FastAPI detail parsing
  • Unify theme picker state and open paths to prevent stale selections and conflicting modal transitions
  • Improve Harbor eval correctness and efficiency with accurate parallel tool-call grouping, persisted effective flags, single-turn ad hoc runs, native result rendering, and batched archive reads
  • Fix and centralize the production telemetry contract

v0.preview.56​

GitHub release

  • Unify model thinking controls as a six-level Effort scale across providers, the CLI, sessions, and eval reporting
  • Redesign /providers and /models with provider-scoped model lists, an effort slider, two-step provider switching, and remembered model preferences
  • Add GPT-5.6 Sol, Terra, and Luna to the OpenAI API and OpenRouter Responses catalogs
  • Wait for loading llama-compatible servers when attaching offline mode instead of reporting them as unavailable
  • Serialize session snapshot writes so older state cannot overwrite newer state
  • Dependency updates

v0.preview.55​

GitHub release

  • Add offline-mode attachment to existing local llama-compatible servers, with broader localhost detection, /v1/models validation, retryable attach errors, UTF-8-safe input, and corrected external-server state handling
  • Add Antix Claude Fable 5 support
  • Add Grok 4.5 to the xAI and OpenRouter model catalogs
  • Fix OAuth provider login started from /providers so the hidden picker no longer captures input or makes the TUI appear frozen
  • Fix Kimi thinking request compatibility by omitting unsupported thinking.type = "enabled" while preserving explicit disabled requests
  • Improve chat composer wrapping so English words wrap at boundaries while long tokens, CJK text, cursor movement, and resizes stay aligned
  • Normalize ambient thinking phrases to sentence case instead of shouting in all caps
  • Compact sessions before the next turn starts to reduce turn-lifecycle and context-management edge cases

v0.preview.54​

GitHub release

  • Add OpenRouter web search request support and forward top_k in OpenAI-compatible requests
  • Update built-in model catalogs
  • Fix multibyte truncation in approval previews
  • Improve settings and credential storage reliability with race-safe ownership and centralized atomic writes
  • Scope permission grants at the permission layer for clearer approval behavior
  • Polish TUI sent-message and composer cursor rendering
  • Clean up the external protocol shape by removing non-wire behavior and decoupling internal storage/logging details

v0.preview.53​

GitHub release

  • Add /goal for goal-driven sessions that keep working until a condition is met, cleared, interrupted, or judged unreachable
  • Improve TUI resize handling to prevent scrollback duplication and content loss, with an opt-in resize_reflow = "purge" setting for terminals that can safely purge and replay
  • Install official pinned llama.cpp prebuilts with SHA-256 verification, atomic versioned installs, GPU-tier selection, and mirror fallback for offline engine setup
  • Improve installer feedback and reliability
  • Refine default prompts for task-contract validation and split Ante product guidance from working defaults

v0.preview.52​

GitHub release

  • Fix skill slash command injection
  • Warn when deprecated tool filter aliases are used
  • Broadly shrink and streamline default prompts and tool descriptions across the agent, skills, OAuth, and TodoWrite flows as model capabilities improve
  • Polish TUI styling for composer input, status line, agent bullets, and sent-message bars
  • Dependency updates

v0.preview.51​

GitHub release

  • Add Claude Fable 5 routes to the model catalog
  • Remove simple agent mode and its dedicated prompt/config path

v0.preview.48​

GitHub release

  • Add an orange grid style for the composer and sent user messages
  • Automatically reconnect on transient mid-stream errors, with consolidated provider error parsing
  • Rework the session into a self-driving actor with a fan-in mailbox, and fix follow-up turn-lifecycle regressions so exactly one end-of-turn event fires per turn
  • Improve offline CUDA detection and local-server diagnostics
  • Read the Zai provider API key from ZAI_API_KEY
  • Fix a UTF-8 truncation bug in the ambient prompt

v0.preview.46​

GitHub release

  • Predict a short, task-specific spinner phrase while you type a longer prompt and show it as the spinner label for that turn (best-effort, runs on the cheapest model, off the critical path)
  • Suggest a likely next prompt as dim ghost text after a turn ends, acceptable with Tab
  • Keep destructive or alarming words out of the predicted spinner phrase
  • Apply a permission-mode change (Shift+Tab) to the in-flight turn, so switching to Auto mid-turn takes effect immediately instead of on the next turn
  • Make MCP and dynamically registered tools obey tool filtering, fixing a bypass where they ignored --allowed-tools/--disallowed-tools; rename the flags to --include-tools/--exclude-tools with the old names kept as hidden aliases
  • Send native web search to the model by default on web-search-capable providers
  • Add gpt-5.4 and gpt-5.4-mini to the OpenAI subscription models
  • Apply ~/.ante/catalog.json as partial provider overlays, patching existing providers field-by-field instead of overwriting them
  • Make skill frontmatter parsing lenient, recovering from common unquoted colons in description: values

v0.preview.45​

GitHub release

  • Add MiMo models via OpenRouter
  • Select flash and default models automatically by weight class (feather/middle/heavy)
  • Infer model vision support from model markers, and return image metadata from Read when the model lacks vision
  • Make native web search a declarative provider flag
  • Rework bash mode to run in the user's shell and reuse the shared crates/exec path
  • Surface settings parse notices in ante doctor
  • Dependency updates

v0.preview.44​

GitHub release

  • Add readline-style line editing with undo/redo to the chat composer
  • Add bash mode (!cmd) for running shell commands inline
  • Make the status line Claude Code-compatible, show the raw context window, and refresh its defaults
  • Lay groundwork for offline mode with a self-contained TUI
  • Make ante-guide read docs from the ante-preview repo checkout
  • Improve auto-memory tool labels in the TUI
  • Handle malformed tool-call arguments cleanly
  • Read the subagent report from the last model message
  • Allow up to 3 max-token continuations per episode
  • Charge model-call usage once per step
  • Restructure the turn model-step loop for readability

v0.preview.43​

GitHub release

  • Add an opt-in auto-memory prompt that teaches the agent to record and recall typed memories across sessions
  • Add OpenRouter GLM 5.2 to the model catalog
  • Clarify max-output-token truncation handling
  • Preserve partially streamed content when a mid-stream error occurs
  • Validate model token budgets when loading the catalog
  • Make tool approval bookkeeping non-fatal
  • Simplify OpenAI-compatible profile policies
  • Tighten the headless check prompt
  • Detect updater installer download failures

v0.preview.42​

GitHub release

  • Add an ante doctor command and speed up startup by decoupling TUI session start
  • Remove the global Ctrl-D TUI exit shortcut
  • Show the Shift+Tab cycle hint in the custom status-line footer
  • Guard background bash commands against trailing ampersands
  • Clean up the offline installer config and release build workflows
  • Dependency updates

v0.preview.41​

GitHub release

  • Rename persisted permission settings to the Claude-style permissions key and drop the legacy permission_settings alias
  • Scope permission rules to known primary tool args (Bash, Agent, Read, Edit) and render TUI tool-call args as the primary value when known

v0.preview.40​

GitHub release

  • Add scoped session permission grants with a subsumption algebra, collapsing batched approval prompts by re-evaluating them against already-granted scopes
  • Cycle permission modes with Shift+Tab
  • Rename the Default permission mode to Strict and drive permission mode through settings and the protocol
  • Rework turn interrupt and steer keys: Ctrl-C interrupts, Ctrl-S steers, and Esc is contextual and also interrupts an active turn
  • Keep the active turn alive across a session or model update
  • Show a live "still working" timer for quiet long-running tools
  • Report context usage as percent used in the status line
  • Cap compaction output by the remaining context window
  • Harden the safe-command classifier for date, file, tree, and abbreviated git options
  • Fix conservative model token budgets
  • Remove unsupported OpenAI subscription models
  • Remove the leading slash glyph from the empty input prompt

v0.preview.39​

GitHub release

  • Add layered permission configuration with an Auto mode and hardened settings handling
  • Stop wildcard permission rules from matching across sequence stages
  • Add a context-window-left metric to the TUI status line
  • Surface a startup notice when settings fail to parse instead of silently ignoring them
  • Add new antix OAuth models
  • Update the model catalog for Anthropic, OpenRouter, and ZAI, and add MiniMax M3 on OpenRouter
  • Stop orphaned subagent turns when a turn is interrupted
  • Classify in-stream OpenAI-compatible error events instead of failing opaquely
  • Improve OpenRouter error detail handling
  • Fix OpenRouter DeepSeek max thinking effort

v0.preview.38​

GitHub release

  • Always stream responses end to end and retire the session streaming flag; suppress raw deltas in headless output
  • Harden Bash safety classifiers
  • Add a command-based status line to the TUI (Claude Code-compatible)
  • Suggest a starter prompt and highlight slash commands on a fresh session
  • Handle terminal stop reasons before retrying empty responses
  • Fix OAuth refresh token response parsing
  • Harden WebFetch against binary responses
  • Fail fast in headless mode when the daemon dies, and explain the stdin EOF wait
  • Extend the typed LLM error taxonomy through streaming and provider error classification
  • Rework the session/turn lifecycle with explicit states and symmetric exits
  • Dependency updates

v0.preview.37​

GitHub release

  • Update the Anthropic model catalog for the latest Claude models
  • Introduce a typed LLM error taxonomy with per-kind recovery hints in error messages
  • Reconnect dropped streams mid-turn and treat cancellation as a first-class turn outcome, including while a stream is being opened
  • Surface content-filter stops like output-limit truncation instead of failing silently
  • Fix Gemini streaming: emit thought text and signature as one thinking delta, resolve tool-response names from the dialog's tool calls, count thinking tokens in output usage, and classify recitation/blocklist stops
  • Harden OpenAI-compatible streaming: accumulate tool-call deltas split across chunks, flush buffered tool calls on premature stream EOF, isolate malformed tool-call arguments, and keep the system prompt when no user message exists
  • Forward max_output_tokens and temperature to OpenAI and preserve truncated output
  • Propagate Anthropic message conversion errors instead of dropping messages
  • Honor the HTTP-date form of Retry-After headers when rate limited

v0.preview.35​

GitHub release

  • Add OpenRouter provider profiles
  • Show a sign-off message and bug-report hint when exiting the TUI
  • Set the terminal window title to "Ante"
  • Handle redacted thinking blocks from the latest Claude models
  • Prevent the Bash tool from inheriting stdin
  • Fix token usage accounting for Anthropic and OpenAI-compatible streaming

v0.preview.34​

GitHub release

  • Add OpenAI-compatible provider profiles
  • Surface subagent activity as live tool updates instead of separate turn events
  • Recover from transient API decode failures instead of crashing the run
  • Allow ANTE_INSTALL_DIR to override the install location and harden the install script
  • Unify the LLM streaming driver across providers for consistent streaming behavior
  • Dependency updates

v0.preview.33​

GitHub release

  • Add ante update --version <V> to pin or roll back to a specific release
  • Retire the legacy latest update channel and transparently resolve it to stable
  • Drive vision/image support from model metadata
  • Reduce Read and multiline Grep latency
  • Fix @-mention handling: dedupe repeats, honor \@ escapes after multibyte characters, and stop dropping large mentioned files

v0.preview.32​

GitHub release

  • Add ante catalog command to print the merged model catalog as JSON
  • Show structured turn errors instead of a raw debug dump
  • Recover from transient connection resets instead of failing the run
  • Fix Anthropic 400 error from unsigned thinking blocks
  • Fix stale OAuth credential cache
  • Migrate the Grep tool to a streaming ripgrep-style search engine

v0.preview.31​

GitHub release

  • Wrap markdown table content in narrow TUI views
  • Fix approval prompt wrapping
  • Use bundled webpki TLS roots for all HTTP clients
  • Use a blocking HTTP client for OTLP telemetry exporters
  • Speed up file searches by pruning VCS directories during traversal

v0.preview.30​

GitHub release

  • Add ante rage command to bundle a bug report
  • Persist tool approvals via "always allow" and store allow/ask/deny rules in settings.json
  • Let Edit create a new file via an empty old_string
  • Suggest a similar path when Edit targets a missing file
  • Handle CRLF files correctly in Read/Edit
  • Harden Edit/Write filesystem guards
  • Allow arbitrary model ids for explicit providers
  • Improve OpenRouter provider defaults
  • Improve responsiveness of grep/glob searches
  • Fix character-based output truncation
  • Fix image decode limits
  • Dependency updates

v0.preview.29​

GitHub release

  • Add Claude Opus 4.7/4.8 and GPT-5.5-pro to the model catalog
  • Drop the retired Gemini 3.1 Flash Lite preview model
  • Add user model overrides to customize built-in model specs
  • Handle malformed user model config entries leniently
  • Use explicit OPENAI_COMPATIBLE_API_KEY for OpenAI-compatible providers
  • Fix persisting of empty sessions
  • Fix status bar truncation and clipping of overflowing hyperlinks
  • Dependency updates

v0.preview.28​

GitHub release

  • Support global ~/.ante/AGENTS.md alongside project AGENTS.md
  • Update OpenAI model catalog and provider selector fallback
  • Add generic LLM model listing across providers
  • Re-enable antix smoke test in release workflow

v0.preview.27​

GitHub release

  • Add OpenAI Responses WebSocket transport (opt-in)
  • Show OpenAI transport in debug panel info tab
  • Improve tool call failure logging
  • Optimize directory tree traversal
  • Tighten OpenAI Response API streaming
  • Normalize blank Grep file type
  • Dependency updates

v0.preview.26​

GitHub release

  • Animate the /compact info block header while compaction runs
  • Show installer download progress
  • Use CDN URLs in release manifests
  • Include provider metadata in session events

v0.preview.25​

GitHub release

  • Add /compact slash command
  • Recover from output-token-limit truncation: keep streamed text and show a hint to send "continue"
  • Auto-compact and retry once when OpenAI requests exceed the context window
  • Fix pager and resume overlays not resizing with the terminal

v0.preview.24​

GitHub release

  • Fix OpenAI subscription streaming requests
  • Fix Unicode clipping in diff viewer
  • Persist update channel overrides, including on install failure

v0.preview.23​

GitHub release

  • Paste images from clipboard with Ctrl+V
  • Add update channel override
  • Log panic crash reports
  • Refactor release artifact publishing and smoke tests
  • Refine Dependabot dependency grouping
  • Dependency updates

v0.preview.22​

GitHub release

  • Add nightly release channel
  • Split stable and latest release channels
  • Fix OpenRouter streaming for thinking (reasoning) parts

v0.preview.21​

GitHub release

  • Add TUI provider selector
  • Simplify model selector
  • Add DeepSeek support for OpenRouter
  • Add random logo variants on startup

v0.preview.19​

GitHub release

  • Improve DeepSeek support
  • Lazy MCP tool registration so daemon doesn't block on warm-up
  • Render MCP tool output as readable text
  • Let background bash survive parent exit
  • Fix public sync messages derivation from tracked paths
  • Fix duplicate auth in public sync
  • Dependency updates

v0.preview.18​

GitHub release

  • Add MCP (Model Context Protocol) support
  • Add browser features
  • Replace BashOutput/KillShell tools with status file
  • Differentiate Bash foreground and background output
  • Add explicit bash background flag
  • Unwrap nested bash -lc wrappers before exec and rule matching
  • Preserve bash output head and tail with mid-omission marker
  • Restore Windows WSL skip and trim bash preview hot path
  • Refactor shell detection handling
  • Move bash tests to integration suite with isolated shell
  • Refine Bash tool description
  • Avoid duplicate shell tool updates

v0.preview.17​

GitHub release

  • Add Windows compatibility
  • Add provider-specific base URL env vars
  • Add extra llamacpp args
  • Update offline models
  • Optimize dialog clone storage
  • Trim ToolEnd shims and dedupe assistant-part emission
  • Wire runtime protocol to shape types and prune protocol shape crate
  • Collapse and tighten protocol helper call sites
  • Fix DeepSeek-v4 interruption bug
  • Fix empty message deletion on interrupt
  • Fix small issues uncovered by DeepSeek testing
  • Fix thinking correspondence
  • Dependency updates

v0.preview.16​

GitHub release

  • Add deepseek-4 model support
  • Update OpenAI and Gemini model presets
  • Split Antix API-key and subscription providers
  • Derive OAuth providers from catalog
  • Make local provider the default
  • Show and preserve current provider in model selector
  • Fix provider fallback resolution
  • Fix sync handling for deleted mapped paths

v0.preview.15​

GitHub release

  • Enable vision for local GGUF models and refresh offline model catalog
  • Fix yolo resume bug
  • Support nested skill metadata
  • Add read-only bash permission heuristic
  • Align headless startup provider handling
  • Move message ID generation into OpMsg/EventMsg constructors
  • Consolidate llm_smoke around session-based tool-call path
  • Split antix into its own catalog module
  • Harden release workflow reproducibility and failure recovery
  • Move thinking option labels into TUI
  • Update connect and model command description

v0.preview.14​

GitHub release

  • Add escape example of Ante and fix config reload bug
  • Fix shutdown bug for offline serve and headless
  • Show changelog on update
  • Support symlinked user skill roots
  • Scope release concurrency by version

v0.preview.13​

GitHub release

  • Add initial Claude Code SDK (agent-sdk)
  • Add offline mode support for headless, serve, and channel modes
  • Add offline mode loading progress bar
  • Promote Evt::UserInput to a protocol-level event
  • Refactor agent-sdk so CLI owns session id
  • Drop redundant search_incomplete field from GrepResult

v0.preview.12​

GitHub release

  • Add --resume CLI flag and exit resume hint
  • Add Slack/Discord integration
  • Add ali-coding-plan builtin support
  • Update log analyzer to accept workflow URL as input
  • Fix Gemini enum problem
  • Improve grep tool: pagination, filtering, glob parsing, count totals, and session cwd resolution
  • Clarify TUI connect command description
  • Remove user group
  • Fix smoke test format
  • Dependency updates

v0.preview.11​

GitHub release

  • Experimental PTY tmux support
  • Update init command description with contextual input
  • Add Gemma4 model
  • Update eval workflow with new harbor
  • Improve offline mode log output
  • Update Antix wirestyle to Anthropic and add Qwen models
  • Adjust offline mode for new llamacpp version
  • Add popular models from OpenRouter
  • Implement explicit update command
  • Dependency updates

v0.preview.10​

GitHub release

  • Update openrouter model name
  • Fix git commit authors for GitHub Action

v0.preview.9​

GitHub release

  • Add dialog snapshot persistence for session restore
  • Add event log persistence and TUI replay on resume

v0.preview.8​

GitHub release

  • Add guide subagent
  • Add number key shortcuts to approval dialog
  • Improve inactive model visibility in model selector
  • Refactor TUI modal state handling
  • Refactor default prompt assembly for agents
  • Update ratatui to 0.30 and tui-input to 0.15
  • Dependency updates

v0.preview.7​

GitHub release

  • Decouple scheduler from review decisions
  • Fix quit bug
  • Update eval workflow and scripts
  • Make browser tool optional
  • Eliminate per-delta buffer cloning in streaming output
  • Deserialize tool results from &Value instead of cloning
  • Sort model selector by current provider first
  • Simplify TUI thinking selector handling

v0.preview.6​

GitHub release

  • Add queued message feature for multi-turn input
  • Add browser tool
  • Fix OpenAI codex backend
  • Reduce tool input cloning
  • Dependency updates

v0.preview.5​

GitHub release

  • Add /statusline command for configurable footer
  • Add PR link status line item
  • Add thinking level selector to model switcher
  • Use theme.secondary for status line text to improve readability
  • Refactor skill module into core/skill
  • Reorganize agent specs
  • Add websocket transport for serve mode
  • Add release skill for tagged releases
  • Fix assistant messages in OpenAI Responses API
  • Dependency updates

v0.preview.4​

GitHub release

  • Add Criterion benchmarking for core fs and Bash tools
  • Add release benchmark baseline reporting
  • Fix update Antix's default URL to public domain
  • Fix typos and spelling
  • Update calculation for benchmarks
  • Move bundled assets to top-level module
  • Dependency updates

v0.preview.3​

GitHub release

  • Prioritize TUI input over protocol events
  • Flatten llm catalog presets
  • Move catalog into llm module
  • Handle queued steers around approval pauses

v0.preview.2​

GitHub release

  • Fix command popup scrolling when selection moves past visible area
  • Add Ante terminus
  • Add standard OAuth support for Antix
  • Fix OAuth callback server cancellation and bind errors
  • Adjust OpenAI reasoning effort mapping
  • Dependency updates