Orca 1.8.5

Fixed

• OpenAI (Codex) sign-in now completes over SSH / on a remote box. The setup wizard used OpenAI's browser OAuth, which redirects to a localhost:1455 loopback the remote box can't receive — the page just kept "loading" and the sign-in never finished. It now uses OpenAI's device-code flow: it shows a short code and auth.openai.com/codex/device, you enter the code, and it polls to completion — no loopback, no copy-pasting a redirect URL. Anthropic keeps its paste-back flow.

Changed

• POST /brain/oauth/:type/start accepts a method (e.g. device_code) so a caller can pick a provider's login sub-flow.

Orca 1.8.4

Fixed

• Complete the 1.8.3 install/setup unification in the source tree — the v1.8.3 tag was missing the file moves (the published npm package was already complete); v1.8.4 has the full, buildable tree.
• orca setup in a non-interactive shell now words its guidance for an interactive terminal and honors --reset; the launcher's "open web UI" prints the URL over SSH instead of claiming a browser opened.

Changed

• Post-review cleanup: removed dead code (an unused step-result field), de-duplicated the cross-platform browser opener between the launcher and the wizard, and simplified slug/id derivation. No wizard behavior change.

Orca 1.8.3

Changed

• orca install now runs the same onboarding wizard as orca setup. The interactive install used to have its own first-run wizard (admin + autopilot + GitHub) that overlapped the setup wizard; there is now a single onboarding path — account, project, AI provider, memory. The autopilot CLI-engine choice and the GitHub PR-workflow prompt live in the web Settings; unattended (flag-driven) installs are unchanged.

Orca 1.8.2

Fixed

• Setup-wizard OAuth no longer looks stuck. The AI-provider OAuth step ran a spinner while showing the authorization URL and the paste-code prompt, hiding them — the sign-in appeared frozen and never surfaced the field to paste the redirect URL back. It now uses the same linear paste-back flow as the web dialog: show the URL (open the browser best-effort), then prompt for the pasted redirect URL / code with nothing competing for the screen.

Orca 1.8.1

First-install onboarding wizard

orca setup is a guided terminal wizard so a fresh npm install -g orcasynth reaches a working setup without hunting through config.

• Five skippable, reversible steps — account, default project, AI provider, memory, review.
• AI provider — an API key, a custom OpenAI-compatible endpoint, or OAuth sign-in (Claude / GitHub Copilot / Codex-OpenAI) with a browser opener and a printed-URL fallback for headless boxes; already-connected accounts are offered for reuse.
• Memory — reuse the AI provider's key with a recommended embedding model (or OpenRouter), validated by the embedding self-test; never blocks on failure.
• Runs where it should — the launcher offers it once on a fresh install (no re-nagging after completion, resume for an interrupted run), and orca setup --reset re-runs it any time.
• Never blocks CI — non-interactive shells (CI / Docker / pipes) just print a next step and exit 0. No postinstall script. All configuration flows through the daemon API.

Install: `npm install -g orcasynth` · then `orca setup`

Orca 1.8.0

Orca is a personal, self-hosted AI agent you talk to — it reasons, calls tools, edits files, runs commands and reaches you across the web dock, the orca CLI, Discord and WhatsApp.

Highlights

Added

• Talk to Orca launcher — running orca in a terminal opens Talk to Orca as the first menu action, dropping you straight into the interactive chat (also orca chat).
• Rewritten user manual — a full agent-first documentation site (getting started, tasks & missions, agents & autonomy, web UI, CLI, brain & chat, plugins, projects, configuration, account & security, architecture), now illustrated with screenshots.
• Plugin illustrations — each bundled plugin ships an illustration on its detail page.
• MCP bridge plugin — connect external Model Context Protocol servers (stdio / HTTP / SSE) and expose their tools to the assistant.
• Configurable agent step limit and per-model context windows (Settings → Orca AI).
• Open a session in the web chat — continue any session in the web dock with full history.

Changed

• Consistent "+N more" pills — one shared pill control across plugin config, cron pickers, user tools and the model catalog.
• Step counter is a stall hint — on Discord and WhatsApp the Step N / MAX line now appears only after ~60s without visible progress and clears on the next tool call or reply.

Fixed

• Friendlier /compact, stripped vision-fallback reasoning leaks, cleaner Discord output, and a round of lint / dead-code / dependency-graph cleanup.

Install: npm install -g orcasynth · Docs: https://orca.dragocz.dev

v1.7.1 — Living dashboard

Living dashboard

The dashboard is reborn as a live operations view instead of a static stat grid.

• Signals row — only meaningful numbers: agents active, decisions waiting, month cost. Muted at rest, toned when they need attention.
• Agent constellation — live agents as pulsing nodes on a sonar backdrop, linked to a central hub; errored or prompt-blocked agents are flagged on their node.
• Live mission card — per-mission progress, running phase, mini terminal preview, changed files, rolled-up cost and lifecycle controls.
• Event stream — the activity feed as short sentences (icon + verb + subject + relative time), newest first.

New endpoint GET /missions/:id/changed-files aggregates each phase's change list into a per-file churn summary.

Visual refresh

Single cycling theme toggle, one-row footer actions, a blue accent, the Quicksand display font, a black logo in light mode, and flatter card/mission surfaces.

v1.7.0

Added

• Light/dark theme support (Light/Dark/System), with a toggle in the sidebar footer. Defaults to following the OS preference and updates live if it changes.
• A language dropdown (English/Čeština) in the sidebar footer, replacing the old inline switcher.

Fixed

• Hardcoded colors across the UI (toasts, terminal, code editor) now adapt correctly to light mode.

v1.6.0

Added

• The dashboard now shows a card with this month's most-used model, total tokens and cost.
• The Stats page has a date-range filter (Today/7d/30d/90d/All/custom), consistent with the Tasks page.

Changed

• Agent/session names throughout the UI now show the friendly name (e.g. "Iris") instead of the internal "orca-Iris" session id.

v1.5.0

Added

• Drag a task card onto another task to relate them: drop onto an epic to make it a phase of that mission, or drop onto a plain task to choose making it a subtask or adding a dependency. Works in the Kanban board and the flat task list.
• The New Task modal now defaults the project picker to the currently active project filter on the Tasks page.