Skip to main content

Changelog

[0.46.0] - 2026-03-24​

Added​

  • feat(voice): real-time streaming voice pipeline with 6-state conversational loop
    • VoicePipelineOrchestrator with LISTENING → PROCESSING → SPEAKING cycle
    • Barge-in detection (hard-cut and soft-fade modes)
    • Three-tier endpointing: acoustic, heuristic, semantic (LLM-based)
    • Speaker diarization with provider delegation and local x-vector clustering
  • feat(voice): 6 streaming extension packs
    • Deepgram real-time STT (@framers/agentos-ext-streaming-stt-deepgram)
    • Whisper chunked STT (@framers/agentos-ext-streaming-stt-whisper)
    • OpenAI streaming TTS (@framers/agentos-ext-streaming-tts-openai)
    • ElevenLabs streaming TTS (@framers/agentos-ext-streaming-tts-elevenlabs)
    • Diarization engine (@framers/agentos-ext-diarization)
    • Semantic endpoint detector (@framers/agentos-ext-endpoint-semantic)
  • feat(speech): SpeechProviderResolver with capability-based resolution, priority chains, and automatic fallback
  • feat(speech): 4 new core providers — Deepgram batch, AssemblyAI, Azure STT, Azure TTS
  • feat(speech): 7 provider extension packs — Google STT/TTS, Amazon Polly, Vosk, Piper, Porcupine, OpenWakeWord
  • feat(voice): Twilio, Telnyx, Plivo telephony providers with webhook verification
  • feat(voice): TelephonyStreamTransport bridging phone calls to streaming pipeline
  • feat(voice): DTMF event support — digits surfaced to LLM context
  • feat(voice): TwiML/XML generation for all 3 telephony providers
  • feat(cli): --voice flags for chat command (STT, TTS, endpointing, diarization, barge-in, port)
  • feat(cli): telephony webhook server and CLI flags

Changed​

  • refactor(voice): streaming-pipeline.ts rebuilt as reusable pipeline handle
  • fix(voice): chat --voice wired with real agent-session adapter
  • fix(session): session.stream() reworked to wrap real sendText() path

[0.45.0] - 2026-03-22​

Added​

  • feat(security): wire 5 guardrail extension packs into SecurityPipeline + SecurityTiers
    • PII Redaction (@framers/agentos-ext-pii-redaction)
    • ML Classifiers (@framers/agentos-ext-ml-classifiers)
    • Topicality (@framers/agentos-ext-topicality)
    • Code Safety (@framers/agentos-ext-code-safety)
    • Grounding Guard (@framers/agentos-ext-grounding-guard)
  • feat(runtime): guardrail packs now active in start/chat with CLI flags
    • --no-guardrails — disable all guardrail packs
    • --guardrails=pii,code,grounding — enable specific packs
  • feat(config): security.guardrailPacks in agent.config.json for per-agent overrides
  • feat(security): 5 security tiers with per-tier guardrail pack defaults
    • dangerous — no guardrails
    • permissive — code safety only
    • balanced — PII + code safety
    • strict — PII + ML classifiers + code safety
    • paranoid — all 5 packs enabled
  • fix: keep the skills tools extension published as @framers/agentos-ext-skills so it stays distinct from the @framers/agentos-skills runtime package
  • feat(api): new sub-path exports: wunderland/config, wunderland/security, wunderland/runtime, wunderland/storage, wunderland/guardrails, wunderland/voice
  • feat(api): barrel exports for config, guardrails, and voice modules

Changed​

  • refactor: guardrail factory functions renamed create*Pack → create*Guardrail for clarity
  • docs: comprehensive updates to all documentation sites

[0.40.0] - 2026-03-14​

Added​

  • feat(cli): voice setup in QuickStart mode — auto-detects OpenAI key, offers 4 provider paths (openai-auto, openai-new, elevenlabs, local)
  • feat(cli): new llm help topic covering 5 providers, switching, OAuth, env vars
  • feat(cli): new faq help topic with 9 common Q&As
  • feat(cli): rewritten voice help topic with full provider coverage (TTS, STT, VAD)
  • feat(cli): voiceVoice, sttProvider, sttModel config fields
  • feat(cli): alias resolution for help topics (ollama/openai/anthropic → llm, faqs/questions → faq)
  • feat: voice runtime with speech catalog and CLI voice commands
  • feat: complete Ollama implementation with hardware auto-detection
  • feat(cli): infinite context window integration in chat loop
  • feat: persistent dedup cache with title similarity for curated picks
  • feat: include skills, deep-research, and more in default extensions
  • feat: local memory tool and updated extension loader
  • feat: add read_document to safe navigation tools
  • feat(cli): overdrive mode, accept-all, tier-aware auth to reduce permission over-prompting
  • feat(cli): --format json for doctor and status commands
  • feat(cli): first-run detection, quickstart, and unified new command
  • feat(cli): standalone tool keys wizard (search, media, voice, devtools)
  • feat(cli): real-time API key validation with retry in setup wizard
  • feat(cli): extensions search subcommand with category filter + categorized listing
  • feat(cli): skills search + recommend subcommands with categorized listing
  • feat(cli): signupUrl/validationUrl for LLM providers + TOOL_KEY_PROVIDERS constant
  • feat(cli): plugins command improvements, system prompt updates
  • feat: cognitive memory config and runtime updates
  • feat: full agency automation — per-agent storage, tool activation, memory pipeline
  • feat: personalized welcome messages with role-based suggestions
  • feat(tool-calling): add deep_research to fallback map
  • feat(discovery): recallProfile (aggressive default, balanced, precision) for discovery tuning
  • feat(runtime): per-turn tool schema narrowing based on discovery results
  • feat(runtime): session.sendText({ toolSelectionMode }) override (discovered | all)
  • feat(runtime): strict tool-name mode via toolCalling.strictToolNames
  • feat(runtime): defensive API key resolution for static/lazy/async inputs

Fixed​

  • fix: correct vector store query result types and remove invalid generic calls
  • fix(cli): prevent partial agent folder on CTRL+C and warn on env key fallback
  • fix: propagate folder-access grants to CLI executor ShellService
  • fix: harden system prompt confidentiality against extraction attacks
  • fix: prevent TUI ghosting when navigating back from views
  • fix: add TUI agents view to prevent crash on "List agents"
  • fix: prevent cross-conversation message history corruption
  • fix(cli): reduce permission prompt noise — session cache + read-only fast-path
  • fix(cli): reduce tool-calling verbosity — gate logs behind debug mode
  • fix(runtime): inject suggestedFallbacks on empty search results + prompt guidance
  • fix: hoist toolExtensions declaration above lazyTools scope
  • fix: resolve TypeScript build errors in storage module
  • fix: use guildMemberAdd with role polling for welcome messages
  • fix(runtime): sanitize/dedupe outbound tool function names for OpenAI-compatible schemas
  • fix(discovery): validate discovery embedding API keys before initialization

Changed​

  • docs: comprehensive tutorials, guides, use cases, and CLI reference for docs.wunderland.sh
  • docs: 5-provider LLM documentation overhaul
  • docs: system prompt confidentiality section in presets guide
  • docs: expanded README and LIBRARY_API.md for adaptive tool exposure

[0.38.0] - 2025-12​

Previous release. See git history for details.