release: call-use v0.1.0 * chore: add PyPI metadata, classifiers, and URLs Co-Authored-By: Claude Opus 4.6 (1M context) * docs: add CLI and MCP sections to README Co-Authored-By: Claude Opus 4.6 (1M context) * docs: add CHANGELOG for 0.1.0 Co-Authored-By: Claude Opus 4.6 (1M context) * ci: add GitHub Actions test workflow Co-Authored-By: Claude Opus 4.6 (1M context) * ci: add PyPI publish workflow on tag Co-Authored-By: Claude Opus 4.6 (1M context) * docs: add CONTRIBUTING guide Co-Authored-By: Claude Opus 4.6 (1M context) * feat: add call-use agent skill for Claude Code and compatible frameworks Co-Authored-By: Claude Opus 4.6 (1M context) * feat: switch default STT from Deepgram to OpenAI Whisper (reduce key dependencies) - Replace deepgram.STT with openai.STT(model="whisper-1") in agent.py - Move livekit-plugins-deepgram from required to optional dependency - Add OPENAI_API_KEY to .env.example, remove DEEPGRAM_API_KEY - Update README, CHANGELOG, and examples to reflect the change Co-Authored-By: Claude Opus 4.6 (1M context) * fix: correct endpoint count in CHANGELOG Co-Authored-By: Claude Opus 4.6 (1M context) * fix: replace deprecated asyncio.get_event_loop() with get_running_loop() Co-Authored-By: Claude Opus 4.6 (1M context) * fix: correct CONTRIBUTING.md anchor link Co-Authored-By: Claude Opus 4.6 (1M context) * fix: move inline imports to top level in server.py Co-Authored-By: Claude Opus 4.6 (1M context) * fix: use user-writable log directory instead of package-relative path Co-Authored-By: Claude Opus 4.6 (1M context) * ci: add ruff lint step to CI workflow Co-Authored-By: Claude Opus 4.6 (1M context) * fix: exclude non-package directories from build discovery Add [tool.setuptools.packages.find] with include = ["call_use*"] to prevent setuptools from treating logs/ as a package, resolving the PackageDiscoveryError during build. Co-Authored-By: Claude Sonnet 4.6 * fix: skill naming, approval guidance, and PyPI-safe links - Rename skill.md to SKILL.md for consistent uppercase convention - Add approval_required note to SKILL.md "When to use" section clarifying MCP runs with approval_required=False; direct sensitive flows to CLI or SDK - Update rule #6 in SKILL.md to explain MCP limitation and correct alternatives - Replace relative example and skill links in README.md with absolute GitHub URLs so they resolve correctly on PyPI Co-Authored-By: Claude Sonnet 4.6 * fix: resolve lint errors and clean up pycache - Remove unused variable `participant` in agent.py - Remove unused import `DispositionEnum` in cli.py (auto-fixed by ruff) Co-Authored-By: Claude Opus 4.6 (1M context) * test: add error path and edge case coverage for 0.1 release Co-Authored-By: Claude Opus 4.6 (1M context) * fix: add voice_id to README, CALL_USE_LOG_DIR to .env.example, error path tests - README SDK example now shows voice_id parameter - .env.example documents optional CALL_USE_LOG_DIR - 14 new tests for error paths and edge cases (108 total) Co-Authored-By: Claude Opus 4.6 (1M context) * docs: improve README for launch appeal Co-Authored-By: Claude Opus 4.6 (1M context) * chore: add py.typed marker and ruff config - Add call_use/py.typed so mypy/pyright recognise the package as typed - Add [tool.setuptools.package-data] to include py.typed in sdist/wheel - Add [tool.ruff] and [tool.ruff.lint] config (py311, line-length=100, E/F/I/W) - Run ruff check --fix and ruff format across call_use/ and tests/ Co-Authored-By: Claude Opus 4.6 (1M context) * fix(agent): remove outbound greeting, add MultilingualModel turn detection - Remove generate_reply greeting from on_enter() that was blocking outbound calls by overlapping with IVR menus. For outbound calls, the callee speaks first and the agent responds via STT→LLM→TTS. - Replace turn_detection="vad" with MultilingualModel() for better turn detection accuracy. - Add turn_detector mock modules to test suite. Co-Authored-By: Claude Opus 4.6 (1M context) * fix(agent): restructure to match official LiveKit v1.4 telephony pattern The agent was not speaking on outbound calls because the session was started AFTER the SIP call was placed. The official LiveKit pattern requires session.start() BEFORE create_sip_participant so the agent's STT/LLM/TTS pipeline is ready when the callee answers. Key changes: - Remove ctx.connect() (rtc_session auto-connects) - Move session creation and session.start() into entrypoint(), before SIP dial - Move SIP creation, participant wait, timeout guard, and disconnect handler into entrypoint() (after session.start) - Replace run() with set_context() for passing LiveKit handles to agent - All existing functionality preserved: state machine, command routing, approval flow, evidence pipeline, DTMF, hang_up, takeover Co-Authored-By: Claude Opus 4.6 (1M context) * Revert "fix(agent): restructure to match official LiveKit v1.4 telephony pattern" This reverts commit 9aba805855c4d2ea711f8d40ad10fcf82d4f522b. * debug(agent): add logging for turn detection investigation Root cause identified: turn detection never fires end-of-turn for SIP audio. STT transcribes correctly but on_user_turn_completed never called. Pipeline doesn't trigger LLM response. Co-Authored-By: Claude Opus 4.6 (1M context) * fix(agent): move models to Agent init per official LiveKit pattern STT/LLM/TTS/VAD now configured in Agent.__init__() instead of AgentSession(), matching the official LiveKit outbound call example. AgentSession is now model-free. on_enter() calls generate_reply() (non-awaited) to start the conversation. Removed unused MultilingualModel import and switched turn_detection to "vad". Co-Authored-By: Claude Opus 4.6 (1M context) * fix(agent): move generate_reply after session.start per issue #2710 - Remove generate_reply() from on_enter() which produces inaudible output (confirmed LiveKit open bug #2710) - Add session.generate_reply() AFTER session.start() in run() method - Move STT/LLM/TTS/VAD from Agent.__init__() to AgentSession in run(), matching the official LiveKit outbound caller example Co-Authored-By: Claude Opus 4.6 (1M context) * fix(agent): resolve speech pipeline — agent now speaks on calls Root cause: generate_reply() inside on_enter() produces inaudible output (LiveKit issue #2710). Fix: moved generate_reply() to after session.start() in run(). Also fixed: call.arguments type changed from dict to str in livekit-agents 1.4.5. Verified: AT&T IVR navigation successful — agent navigated menu, answered questions, connected to human representative "Mike". Co-Authored-By: Claude Opus 4.6 (1M context) * fix(agent): capture agent speech transcript, clean up debug logging - Replace non-existent "agent_speech_committed" event with "conversation_item_added" (the correct v1.4.5 event), filtering on role == "assistant" to capture agent speech in transcript - Remove debug log lines (on_user_turn_completed info, agent speech debug print) while keeping functional logging - Clean up send_dtmf_events arguments handling: simple isinstance branching instead of try/except JSON parse - Fix unused variable (participant) and line-length lint violations - Reformat BASE_PHONE_INSTRUCTIONS as implicit string concatenation to satisfy E501 line-length rule Co-Authored-By: Claude Opus 4.6 (1M context) * test(agent): add comprehensive BDD tests for all call-use scenarios Adds 81 BDD-style tests across 10 test classes covering call lifecycle, IVR navigation, approval flow, human takeover, evidence pipeline, phone validation, CLI behavior, MCP server, data message routing, and instruction building. Co-Authored-By: Claude Opus 4.6 (1M context) * test: add user journey tests covering all user-facing scenarios 28 tests simulating real user workflows: import, CLI help, error messages, SDK validation, CLI exit codes, MCP non-blocking design, and REST API creation. Co-Authored-By: Claude Opus 4.6 (1M context) * fix: resolve release audit findings — deepgram dep, docs accuracy, approval UUID - H1: Move livekit-plugins-deepgram from optional to required dependencies - D1: Update README/CHANGELOG from "OpenAI Whisper STT" to "Deepgram Nova-3 STT" - M3: Replace class-level _approval_counter with uuid4 for thread safety - D3: Add DEEPGRAM_API_KEY to .env.example Co-Authored-By: Claude Opus 4.6 (1M context) --------- Co-authored-by: Claude Opus 4.6 (1M context)

Uh oh!

feat: Caller ID ownership verification via Twilio Lookup API #6

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions