3dbbf63fb9cd721669a99ea2b00bbbccfdf2ff71
11
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Bohdan Triapitsyn
|
fcbe2c0414 | fix(settings): clean up orphaned temp files and harden atomic writes | ||
|
Bohdan Triapitsyn
|
b55152db6f | fix(usage): refresh work status quotas automatically | ||
|
Bohdan Triapitsyn
|
34d0ff7383 |
feat(walkthrough): guided AI walkthrough for diffs, branches, and PRs (#2572)
A diff is ordered by file path, which is almost never the order in which a change makes sense. This adds a Walkthrough surface that reorders it: the model groups related hunks into stops, explains what each group changes about behavior, and orders the stops so each builds on the last. It explains and orders; judging code stays with the existing Review action. Reviews uncommitted work (all, staged, unstaged), a branch against its base, or a pull request. Generation is always user-initiated — nothing runs on a timer, on a file change, or as a side effect of opening a panel. Invariants worth preserving: - Hunk identity is derived on the server and only there. Ids are content hashes, so an anchor that no longer resolves is proof the code it described changed, and staleness needs no heuristics. The client matches ids to ids and never recomputes them; two implementations would have to agree forever. - The digest is never truncated. A diff that does not fit the model's context is refused with an actionable reason, because a walkthrough written against half a diff reads as confident and is wrong. - Nothing disappears. Lockfiles and other generated output are excluded from the model's input by name — never by size — and everything no stop covers is listed at the end, so "have I seen all of it" stays answerable. - Cost is explicit. Results are content-addressed, so returning the working tree to an earlier state costs nothing; generation outlives its request, so a refresh detaches the client rather than discarding paid-for work, and only an explicit cancel stops it. Supporting changes to shared modules: - git: expose the existing getRangeDiff as GET /api/git listUntrackedPaths and getUntrackedDiffs. The latter resolve the repository once for a batch instead of per file, taking a panel ~340ms on an 80-file working tree. - small-model: structured output across four wire forma and abort signal, and an onOverflow policy so an oversized prompt fails loudly instead of being silently clipped. A provider remembered so the prompt-side fallback goes first next time. - models.dev metadata: surface structured_output as tri false blocks a model, a missing field does not, because the catalog omits it for roughly half of all models. Desktop and tablet only: VS Code serves Git through its these routes, and the mobile shell does not consume the surface registry. Docs: packages/docs walkthrough page in English and all eight locales. |
||
|
Bohdan Triapitsyn
|
bb45164ae8 |
feat: session goals - server-driven goal loop with independent small-model audit (#2148)
Arm the target button in the composer and the next prompt becomes a goal: the server keeps the session working toward it (idle tick -> small-model audit -> continuation) until the objective is verifiably complete, blocked, or out of budget — even with the UI closed. Server (packages/web/server/lib/session-goal): - event-driven loop on the global SSE hub; goal state lives in session.metadata.openchamber.goal (merge-safe patches, stale-write guard by goal id), so it survives restarts and syncs to every client for free - the small-model audit (objective + last assistant turn only, language pinned to the objective) is the sole termination authority; blocked needs 3 consecutive verdicts, audit outages tolerate one unaudited continuation then stop the goal as resumable-blocked - hard stops: optional token budget, auto-continuation cap (Resume grants a fresh allowance), turn errors; user abort pauses the goal instead of blocking it, and resuming over an aborted tail nudges immediately - token accounting as a snapshot of the latest turn (input + cache.read + output), goal-relative via a creation baseline and segmented across compactions; a compaction summary skips the audit and continues - continuations reuse the session's own provider/model/agent/variant UI: - three-mode target button (arm / disarm / manage dialog), informational goal strip with inline pause/resume and an Evaluating indicator, sidebar state glyph, objective length counter (2000-char server clamp), read-only completed goals - goal entry points: composer (sessions and drafts), start-new-session- from-answer dialog, plan implement dialog (plan content becomes the objective), scheduled tasks (Run as goal + budget) - Settings -> Chat -> Goal: feature toggle + default token budget with three-layer parity (web server, client persistence, VS Code bridge); VS Code renders goal state but hides the entry points (the loop runs in the web server only) Notifications: per-turn "ready" notifications are suppressed while a goal is active; settling sends one final notification (desktop, web-push, APNs generic titles with the session name as body) honoring the completion toggle. Error/question/permission notifications are untouched. Docs: user guide (session-goals) in all 9 locales + sidebar entry, scheduled-tasks cross-reference, server module DOCUMENTATION.md. |
||
|
Bohdan Triapitsyn
|
182fc89b8d | feat: split session recap and suggestion settings | ||
|
Bohdan Triapitsyn
|
28f0736d69 |
feat: small-model utility calls on existing OpenCode providers (#2049)
Adds a server-side "small model" capability: direct, cheap LLM calls that reuse the user's existing OpenCode provider logins — the mechanism OpenCode uses internally for titles and summaries but does not expose through the SDK or plugins. Zero new dependencies; plain fetch with per-provider wire formats, credentials never leave the server. Core (packages/web/server/lib/small-model): - Resolution mirrors OpenCode's session scoping: explicit settings override → small_model from the OpenCode config → family scan within the session's provider → the session's own model. The global provider scan only serves callers without a session context, and background callers forbid it entirely (restrictToPreferredProvider), so conversation content never reaches a provider the user didn't pick — explicit choices excepted. - Per-provider auth replicating OpenCode's plugin loaders: GitHub Copilot (device token as bearer, no exchange), ChatGPT plan via the codex Responses API (single-flight OAuth refresh written back to auth.json), Anthropic messages, Google generateContent, generic OpenAI-compatible. - OpenCode's free models (opencode/big-pickle, *-free) are never called directly; unauthenticated providers are skipped by design. - Prompt clamping to the model's catalog context limit; thinking disabled where a wire switch exists (Z.AI/GLM, MiniMax-M3, Gemini Flash); robust content parsing with a clear error when a thinking model spends its whol budget on reasoning. - Settings → Sessions gains a Small Model group: use-default checkbox plus an override picker limited to authenticated providers, persisted with web/desktop/VS Code sanitization parity. Consumers: - Session assist: a server-side watcher on the global SSE hub generates a short recap and one suggested follow-up after a session idles quietly fo a minute, stored on session metadata (openchamber.assist). Freshness is keyed to the last assistant message id, so new activity invalidates the payload everywhere with no extra writes. The chat shows the recap under the last message after five quiet minutes and the suggestion as a dismissible chip above the composer (tap fills the input, never sends). Gated by a new Chat setting (default on) that is a hard generation switch. Language is anchored to the conversation itself, with a script-mismatch guard against model/backend language hallucination. - TTS: a third input mode, summarized — long replies are condensed to spoken prose before playback on any TTS engine. - Git: commit-message and PR generation moved off the active chat session onto the small model fed with real diffs and the commit list (bodies included), with a session-transport fallback for free-model-only setups. - Notes: Add to notes distills long selections into 1-3 dense sentences preserving exact identifiers, with verbatim fallback on failure. Fixes along the way: - The global event watcher now starts unconditionally; it was gated behind the desktop-notify env, leaving the server-side event hub dead in packaged apps. - OpenCode re-emits message.updated for old user messages after idle; the watcher no longer mistakes those for new activity. - Session metadata merges from a fresh read right before the PATCH, so writes made during the generation window (suggestion dismissals, review links) are preserved; the assist runtime stops during graceful shutdown. |
||
|
Erman HAVUÇ
|
684d55f4aa |
feat(usage): add toggle to hide prediction rows on usage cards (#1420)
* feat(usage): add showPredValues setting to quota store * feat(usage): register usageShowPredValues in settings sanitizers * feat(usage): add i18n key for show predictions toggle * feat(usage): add show predictions toggle to sidebar * feat(usage): hide pred row by default, gate on showPredValues * fix(usage): gate header dropdown PaceIndicator behind showPredValues * fix(usage): gate VSCodeLayout PaceIndicator behind showPredValues * fix(usage): correct indentation drift in Header.tsx PaceIndicator blocks |
||
|
jkker
|
1663185a75 |
fix installed skills discovery and improve editor UX (#1296)
* fix skills discovery from opencode * Fix stale skill description after frontmatter removal * fix: align vscode skill discovery parity |
||
|
Bohdan Triapitsyn
|
dcd68baa09 |
fix: cross-client settings sync and sidebar session pagination
- Paginate /experimental/session via time.updated fallback so sidebar loads past the 200-item first page when server omits x-next-cursor. - VSCode extension persists all settings in shared ~/.config/openchamber/settings.json (not only opencodeBinary), matching Desktop and Web. Canonical read from disk with globalState fallback for eager migration of pre-existing users. - Desktop settings sync: guard the contextBridge read-only __OPENCHAMBER_HOME__ assignment that was silently throwing a TypeError and skipping the whole applySettings chain — leaving server-synced values (autoDeleteAfterDays, autoDeleteEnabled, sessionRetentionAction) stuck at local defaults. Also wait for Zustand persist hydration before applying server settings to avoid overwrite races. |
||
|
Bohdan Triapitsyn
|
5f0d1623ae |
feat: expand magic prompts coverage and split generation prompts (#835)
- Add configurable visible/instructions prompt families for commit/PR generation, PR checks/comments flows, and git conflict resolution helpers. - Refactor prompt sending to explicit visible + synthetic parts instead of newline-based splitting, with legacy override migration for old keys. - Polish Magic Prompts settings UX with grouped sidebar entries, tooltip-based descriptions, AI icon, and validation that visible prompts cannot be empty across web and VS Code runtimes. |
||
|
Bohdan TriapitsynandIuliia Ivashko
|
c9e31a0e6c |
perf: harden sync architecture and modularize runtimes (#803)
* fix: added desktop app background throttling
* perf: add streaming debug metrics panel
- Show streaming performance metrics in the debug panel
- Auto-enable stream profiling while the panel is open
- Add JSON export for sharing UI and VS Code metrics
* perf: batch streaming updates more aggressively
- Buffer message deltas and metadata updates to cut render churn
- Skip no-op part updates before they touch the message store
- Fix the desktop debug panel shortcut binding
* perf: split streaming event handling and coalesce deltas
- Move streaming content events onto a dedicated fast path
- Defer non-critical stream side effects off the hot path
- Merge repeated message delta events before they reach the UI
* perf: isolate streaming rows from chat rerenders
- Memoize chat rows against render-relevant message changes only
- Read live assistant text directly from store to narrow streaming updates
- Split the active streaming entry from the stable message list path
* perf: streamline chat streaming and SSE proxying
- Reduce chat rerenders around the active streaming path
- Simplify server SSE forwarding to avoid duplicate proxy work
* fix: preserve the first streaming text chunk
- Show the initial text chunk immediately before batched deltas arrive
- Bypass batching for the first text or reasoning part update
- Keep later streaming updates buffered for performance
* perf: align streaming/render hot paths with opencode parity
* perf: harden turn/cache stability and stale delta suppression
* fix: stabilize chat rendering and disable timeline interactions
- Disabled timeline dialog access from shortcuts, commands, and chat input
- Reduced chat render churn by simplifying message list and turn staging behavior
- Improved session-switch stability to prevent update-depth crashes
* perf: track static message rerenders during streaming
* perf: reduce sorted-mode activity rerender fanout
* perf: reduce chat rerender fanout and add active-turn metrics
- Reduced sorted-mode rerender coupling by tightening turn context propagation
- Added a metric for static rerenders outside the active turn during streaming
- Exposed new chat render counters in the debug panel for parity tracking
* fix: keep sorted activity mounted while stream grows
* fix: stabilize session and history scroll rendering
* refactor: decouple server routes from index
* refactor: extract fs module from server index
* refactor: move opencode route ownership into module
* refactor: extract notification route registration
* refactor: extract opencode and notification runtimes from index
* refactor: extract settings runtime and complete server modularization pass
* refactor: modularize server config, skills, icons, and tunnel routes
* refactor: extract server modules from monolithic index.js
Split proxy, routes, runtime helpers, and notification emitter
into dedicated modules under packages/web/server/lib/.
* refactor: replace session/message stores with SSE-driven sync layer
Delete ~9200 lines of old architecture (useEventStream, messageStore,
sessionStore, useSessionStore, questionStore, useTodoStore, client SSE).
New sync layer: event pipeline with coalescing + 16ms flush, pure event
reducer, per-directory child stores with LRU eviction, cursor pagination,
optimistic updates, deferred timeline staging, text throttle.
Migrate all UI consumers to sync hooks (useSessionMessages,
useSessionMessageRecords, useSessionStatus, useSessionPermissions, etc).
Strip session-ui-store to UI-only state, delegate SDK ops to
session-actions with abort-if-busy, optimistic store updates, and
response merging for revert/fork/archive/delete.
Add notification-store for SSE-driven session attention tracking,
cross-directory GlobalSessionStatusStore for sidebar indicators,
client-side diff snapshot sanitization to prevent memory bloat,
and revert message filtering via useVisibleSessionMessages.
* feat: notification store, session actions, activity detection
Add notification-store.ts for SSE-driven attention tracking.
Add sanitize.ts to strip diff snapshot memory bloat.
Add session-actions.ts with optimistic revert/fork/archive/delete.
Improve useSessionActivity with incomplete-message fallback.
Delete useServerSessionStatus polling hook.
* fix: add directory param to all SDK calls, fix command/shell/abort routing
All SDK calls in session-actions.ts now pass directory parameter —
required by OpenCode server to scope session operations. Without it,
abort, commands, revert, fork, and other operations returned 500.
Add routeMessage() in session-ui-store for shell mode (session.shell),
slash commands (session.command), and normal prompts. Command lookup
checks both sync child store and useCommandsStore. Handle /compact
locally via session.summarize().
Implement getContextUsage() to restore header context usage display —
reads token counts from last assistant message in sync store.
* refactor: replace custom API proxy with http-proxy-middleware
Remove ~280 lines of custom proxy code: forwardSseRequest,
forwardGenericApiRequest, collectRequestBodyBuffer, header
manipulation, hop-by-hop filtering, SSE block buffering.
Replace with single createProxyMiddleware() call that handles
SSE streaming, large bodies, and timeouts out of the box.
Dynamic router for OpenCode port changes after restarts.
Auth headers injected via proxyReq hook.
Keep: readiness gate, Windows session merge, API prefix detection.
* perf: targeted event draft cloning to fix streaming render cascade
Event handler was eagerly cloning all state slices on every event,
breaking Zustand selector referential equality. During streaming
(~60 events/sec), this caused every subscriber to re-render regardless
of which slice actually changed.
Now only clones fields the specific event type mutates. Also extracts
StatusRowContainer to isolate high-frequency useAssistantStatus
subscription, removes dead messageStreamStatesMap subscription from
ChatContainer, and narrows useAssistantStatus to only track last
assistant message parts.
MessageList renders: 1972 → 296 per streaming session (-85%).
* fix: null safety for sync state slices
Add defensive ?? {} guards on permission, question, session_status,
and message record access. Prevents crashes when child store state
is partially initialized during bootstrap.
* perf: dedup inflight SDK calls, extract concurrency util, delay PR tracking
Extract mapWithConcurrency to shared lib/concurrency.ts. Add in-flight
dedup for loadProviders/loadAgents to prevent concurrent duplicate SDK
calls. Delay initial PR background tracking by 5s to reduce startup
CPU burst.
* fix: header session lookup across all child stores
Session title and context panel click failed when session belonged to
a different directory than the current child store. Fall back to
getAllSyncSessions() to search all initialized stores.
* chore: bump @opencode-ai/sdk to 1.3.5
* docs: add sync event handling guide
* Optimize session prefetch and improve delete/archive UX
- Add settlement delay to session prefetch to avoid race conditions on
rapid session switches
- Reduce git diff prefetch and session cache limits for better performance
- Implement optimistic UI updates for session delete/archive operations
with proper rollback on failure
- Wire session prefetch hook into SessionSidebar with sync integration
* Add file content cache and sync optimizations
- Wrap FilesAPI with in-memory LRU cache for file content with dual
constraints (entry count and byte size)
- Optimize chat timeline scroll restoration using useLayoutEffect
- Preserve React references in message and part arrays to prevent
unnecessary re-renders when prepending history
- Add session prefetch TTL cache to prevent redundant fetches
- Integrate session prefetch cache clearing with eviction flow
* Improve session sidebar error handling and add diff prefetch filtering
Load active and archived sessions independently using Promise.allSettled
to prevent one failure from blocking the other. Add retry logic to session
API calls and skip large files during diff prefetch to improve performance.
* Replace sendMessage with optimisticSend wrapper
Introduces optimistic UI updates for normal chat messages to provide
instant feedback. Messages appear immediately in the UI while the API
call executes in the background, with automatic rollback on errors.
* perf: split stores, proper optimistic send, fix revert/directory bugs
- split session-ui-store into voice/input/selection/viewport stores
to reduce subscriber re-evaluation during streaming
- wire optimisticSend through useSync shadow Map infrastructure
matching OpenCode's pattern (no heuristic part detection)
- port OpenCode Identifier.ascending ID format for correct sorting
- pass messageID to promptAsync to prevent duplicate messages
- fix worktree directory not propagating to session actions
(dynamic dir() via opencodeClient.getDirectory)
- fix setCurrentSession accepting directoryHint for new sessions
- fix revert not hiding messages (session limit was 5, bumped to match loaded count)
- fix revert optimistic message removal from store
- fix load-more flicker (useLayoutEffect scroll compensation)
- add prefetch TTL cache, file content LRU cache
- add session prefetch for adjacent sessions
- add instant archive/delete (optimistic before SDK call)
- migrate legacy window.__zustand_session_store__ to session-ui-store
- add retry + independent error handling for archived sessions
- add AGENTS.md performance rules
* perf: startup optimization — dedup, caching, light git status, diff rendering gates
- defer diff prefetch to git tab open, reduce concurrency 4→2, skip >500 changed lines
- cap project git checks concurrency (2), directory status probe (3)
- dedup provider/agent loading, github auth, worktree list (in-flight + TTL caches)
- delay PR tracking 5s, cache 403 search failures per-repo
- coalesce settings PUT (200ms debounce), cache settings GET (2s TTL)
- cache canonical directory resolution (60s TTL)
- persist missing directory status to localStorage (10min TTL)
- light/heavy git status: polling skips numstat+line counting+rev-list
- large diff rendering gate (>500 lines → "render anyway" button)
- tokenization degradation for >500KB files in Pierre
- parallelize main.tsx pre-render awaits
- batch sidebar file tree expanded paths restoration (3 at a time)
- remove bare useConfigStore() subscription in AgentsPage
- sync worktree sandboxes to OpenCode SQLite DB
- fix RightSidebarTabs ternary → explicit tab matching
- defensive guards on sync state (session_status, permission, question, message)
* fix: add defensive guards on remaining sync state field accesses
guard session_status, permission, message, todo, part, config with ?? {}
in useDirectorySync selectors, session-cache, and bootstrap
* fix: add missing directory dep to useCallback in use-sync.ts
* fix: preserve diffStats when light-mode polling overwrites status
* perf: optimize startup git status polling and diff rendering
Preserves diff stats when lightweight polling updates repository status
Reduces startup overhead with smarter git polling and store updates
Adds detailed optimization and migration docs for next performance steps
* fix: keep chat diff stats stable during git status updates
Prevents lightweight git polling from dropping diff statistics
Keeps MessageList diff indicators consistent while status refreshes
Improves reliability of git-aware chat rendering
* fix: user animation replay, queued message variant, startup provider loading
- consume animation ID after first play to prevent re-animation
on neighbor assistant message completion
- capture send config (model/agent/variant) at queue time matching
OpenCode's FollowupDraft pattern instead of re-resolving at send time
- replace one-shot startup recovery effect with polling interval
that retries every 2s until providers and agents load
- fix optimistic bridge to avoid re-render loop (stable ref wrappers)
* chore: update tauri to 2.10.3 and all plugins to latest
- tauri 2.9.4 → 2.10.3
- tauri-build 2.5.3 → 2.5.6
- tauri-plugin-dialog 2.4.2 → 2.6.0
- tauri-plugin-log 2.7.1 → 2.8.0
- tauri-plugin-shell 2.3.3 → 2.3.5
- tauri-plugin-updater 2 (floating) → 2.10.0 (pinned)
- @tauri-apps/api ^2.9.0 → ^2.10.1
- wry 0.53.5 → 0.54.4 (transitive)
* refactor: decouple web server index orchestration runtimes
* fix: align VS Code runtime behavior with web and reduce draft view CPU load
- Queue VS Code bridge and SSE startup requests until API readiness to avoid false bootstrap failures
- Make agent manager actions directory-aware and remove real worktrees with safer partial-failure handling
- Replace heavy logo animation path with a lightweight pulse to cut draft-session CPU usage
* fix: restore auto-selected file sending in chat input
- Send server-selected files as proper file URLs in the message payload
- Include server-backed attachments in submit flow instead of dropping them
- Restore queued-message attachments through the refactored input store
* fix: restore session model selection consistently on session switch
- Restore agent, model, and variant from the latest loaded user message for each session
- Wait for session messages before applying restored selections to avoid stale or missing state
- Remove legacy session-choice inference paths that caused overlap and instability
* fix: restore permission replies and auto-accept across sessions
- Scope permission and question replies to the target session directory so answers take effect reliably
- Make permission auto-accept immediately handle pending requests and react to new permission prompts
- Keep parent-session handling working for child-session requests through the shared response path
* feat: add reusable fuzzy branch fuzzy-search helper and dialog integration (#798)
* feat: add reusable fuzzy branch search for worktrees
* chore: drop planning docs from feature branch
* feat: make worktree branch refresh manual
* feat: add configurable session retention action
* refactor: centralize global session state in ui store
* fix: cancel debounced permission push after reply
* docs: clarify global and directory session store architecture
* docs: refine agent development rules and session activity guidance
- Clarify agent code of conduct and durable development patterns
- Add explicit shared-store rerender and live-state guidance
- Narrow session activity fallback to avoid stale working state
* chore: updated .gitignore
---------
Co-authored-by: Iuliia Ivashko <yulia.ivashko@gmail.com>
|