ORBIS v0.2.174 — cleaner tool acknowledgements

ORBIS v0.2.174 — cleaner tool acknowledgements

• Suppress a tool's opening acknowledgement when the agent has already spoken a preamble or answer in the same turn.
• Drop queued opening acknowledgements after an interruption or new user turn.
• Keep one acknowledgement for tool-only turns, with streamed answers, progress updates, and explicit deliveries preserved.

Public beta for Apple Silicon Macs.

ORBIS v0.2.173 — more natural conversation

ORBIS v0.2.173 — more natural conversation

• Shorter, more conversational replies: one short sentence by default, with more detail when you ask for it.
• Remove Markdown formatting before speech synthesis across all voice backends, including streamed answers and direct announcements.
• Keep saved personas' character while applying the same concise spoken style, without routine plan narration or follow-up offers.

Public beta for Apple Silicon Macs.

v0.2.172

ORBIS v0.2.172 continues the Apple Silicon Mac public beta with native audio recovery reporting, structured delegation progress, and restored opt-in wake activation.

• Restore opt-in wake-word activation with verified model downloads, real model warmup, visible failure state, and hard-mute protection. Tap to talk remains the default; Hey Jarvis is the fresh phrase preference and Hey Orbis remains experimental.
• Reaccept native audio socket connections, discard stale disconnected audio, detect stopped capture, and show explicit relaunch guidance after capture or backend loss.
• Keep structured delegation progress, questions, cancellation, and terminal task results aligned with the authoritative protoAgent task lifecycle.
• Prevent stale wake snapshots and detector generations from reopening listening after native audio loss; preserve activation edits made during model downloads.
• Fix hook ordering in orb settings and improve mobile marketing navigation, onboarding, privacy explanations, and Mac download guidance.
• Verify the candidate with 1,285 passing Python tests (2 optional skips), 48 native Rust/model tests, 56 frontend tests, production builds, and bundle integrity checks.

This is a public beta. Live microphone, spoken wake recognition, device switching, sleep/wake, and long conversation acceptance remain unperformed and are tracked in #743. Automated model checks establish loadability, not spoken recognition accuracy. ORBIS does not automatically restart failed hardware streams or the sidecar; use the recovery action shown in the app. The stock wake models carry CC BY-NC-SA 4.0; model source and license are displayed in settings. Team review reported no blockers or majors, with incomplete structural review coverage tracked in #741. Remaining review follow-ups are #733–#744 and #491.

v0.2.171

v0.2.171

• fix(release): support empty PyApp cargo args on Bash 3.2 (#726)
• chore(changelog): v0.2.170

v0.2.170

v0.2.170

• fix(distribution): verify PyApp UV bootstrap archive (#724)
• fix(build): isolate local sdist frontend (#723)
• fix(ui): separate app and voice readiness (#717)
• docs: align desktop access sidebar label (#721)
• fix(macos): keep tray app discoverable (#720)
• fix(voice): serialize cold model warmup before Pipecat startup (#715)
• perf(desktop): use pinned UV for PyApp installs (#716)
• chore(changelog): v0.2.169

v0.2.169

• The local hub endpoint is promoted as the default agent connection point.
• Hub boot availability is hardened to reduce failure modes during startup.
• Hub startup confirmation is bounded to prevent indefinite blocking.
• A warning is emitted if hub startup does not finalize within the expected window.
• Web boot status events are closed to resolve race conditions during initialization.

Release Integrity

• macOS DMG artifacts are validated against authoritative sources to ensure integrity.
• The macOS artifact signer is pinned to prevent unexpected signature mismatches.
• Automated version bumps are kept in sync to prevent drift between release channels.

v0.2.168

v0.2.168 hardens voice reliability, makes A2A delegation durable across sessions, and adds observability for the new delegation path.

Voice & Delegation

• Verbal cancel phrases now cancel live delegated tasks, clearing pending input-required questions and marking local handles canceled.
• When an A2A task parks on input-required, ORBIS routes your next utterance as the answer to that task, preserving task ID and context.
• Reconnecting to a session requeries all live outbound tasks, delivering results missed while away exactly-once and re-arming input-required questions.
• The /a2a/callback endpoint now correlates push-backs by task ID to mark durable rows terminal, preventing duplicate delivery on subsequent requeries.
• A new durable outbound-task registry persists task handles in memory schema v5, surviving barge-ins, disconnects, and restarts.
• Loopback A2A delegates now register push notification URLs, allowing results to be pushed to /a2a/callback even after dispatch timeouts.
• The presence loop now yields after two spoken check-ins instead of narrating indefinitely, speaking one explicit yield line before going quiet.
• A dead-turn recovery line now plays when an LLM completion emits no text or function calls, replacing silence on truncated or empty responses.
• Reasoning tags are stripped from streamed LLM text before TTS, preventing leaked reasoning blocks from being spoken aloud.

Performance & Observability

• Every delegation dispatch now generates a Langfuse span and success/error counters, with A2A trace headers stamped on outbound requests for end-to-end trace stitching.
• Session tool logs are now persisted to sessions.tool_calls, recording up to 500 function calls per session.
• The micro-LLM filler tier now has a 300-token budget and 4.0s timeout, fixing silent failures when reasoning models consume the token allowance.
• Canned fallback lines now play when micro-LLM progress generation fails or times out, ensuring the presence loop never falls into dead air.

Distribution

• Docker runtime is verified before publish, and exact Docker image imports are gated.
• Docker publication is now bound to verified refs, isolating PR Docker verification permissions.

v0.2.167

ORBIS now supports running directly on Claude Pro/Max or ChatGPT/Codex subscription plans via integrated OAuth sign-in, replacing API key requirements for primary voice and text workflows.

Subscription Provider Support

• Anthropic OAuth and OpenAI Codex providers integrate into the LLM pipeline, supporting in-app sign-in, automatic token refresh, and credential lifecycle management.
• Voice sessions boot signed-out rather than crashing if credentials are missing, allowing the system to prompt for sign-in instead of failing session setup.
• Out-of-pipeline text agents, filler generation, and drift analysis now route through OAuth providers, preventing failures on subscription-only accounts.
• The LLM test endpoint and model fetching respect the active OAuth provider, ensuring accurate diagnostics without manual API key entry.

Credential Management

• OAuth status displays precise expiration timestamps, refresh capability, and credential durability, distinguishing between managed stores, borrowed CLI logins, and static environment tokens.
• macOS Keychain acts as a fallback credential source for Claude Code, with a 30-second cache preventing subprocess overhead on high-frequency model calls.
• Identity system prompts are strictly formatted to match Anthropic’s byte-exact requirements, resolving false rate-limit errors caused by malformed prefix blocks.

v0.2.166

This release updates the drawer interface, corrects version tracking on the download page, and clarifies first-run expectations.

Application UI

• The drawer footer displays the current version and a linked studio wordmark across every tab.
• The studio link routes through the system browser via IPC, keeping the app interface intact.
• The version string reads directly from the Tauri build configuration instead of a hardcoded constant.

Release & Onboarding

• The download page syncs the latest release instead of silently serving a stale version.
• First-run documentation clarifies the multi-minute initial unpack and the required language model setup.

v0.2.165

Voice Reliability

• Bounds degenerate tool loops on the voice path to prevent infinite spinning on identical requests.
• Forces speech synthesis after three repeated tool-call round-trips to eliminate dead air.
• Applies escalating nudge at two repeats before disabling tool calling entirely at three.
• Configures the brake per backend, using tool_choice: "none" for OpenAI-compatible APIs and schema withholding for Ollama and MLX.

Tool Loop Policy

• Introduces agent/tool_loop.py as a pure, testable policy module with no Pipecat dependencies.
• Supports environment-tunable thresholds via VOICE_TOOL_LOOP_NUDGE_AT and VOICE_TOOL_LOOP_STOP_AT to match existing iteration caps.
• Uses greater-than-or-equal comparisons for thresholds because request-only edits do not persist notes in history for the model to reference.
• Pins tests across all four backends to ensure the guard is reached and does not rot in isolation.