ContractSpec
ContractSpecOpen spec system
Menu

Choose your entry point

ProductTemplatesPricingDocsChangelogGitHub
Start with OSSExplore Studio
Back to changelog index

12.0.1

May 25, 2026 · 3 packages · 4 unique changes · 1 release entry

integrationslibs

This release affects the integrations, sharedLibs familyies.

Run contractspec connect adoption resolve --family integrations to see how it impacts your project.

Release summaries

  • m3-ai-agent-graph-ops

    M3 — Four graph AI LLM tools (graph.search, graph.filter, graph.cluster, graph.suggest) registered in ai-agent tool registry; two new voice input integration packages (voice-input-web, voice-input-native).

    maintainer

    Four LLM graph tools are registered via the tool registry. All four require human confirmation before applying suggestions. useAiCostGuard tracks per-session spend. Voice input packages are new published packages.

    integrator

    Wire graph AI tools via the ai-agent tool registry. Use createWebVoiceInput for web STT and createNativeVoiceInput for native STT. Never hardcode Deepgram API keys; pass via config. Native voice requires a custom Expo dev client.

    customer

    AI graph operations (search, filter, cluster, suggest) are available with explicit confirmation before any change is applied. Voice input supports microphone and paste-mode fallback.

Upgrade steps

  • Register and wire graph AI tools

    manual

    Add graph.search, graph.filter, graph.cluster, and graph.suggest to the agent tool registry for graph surfaces.

    Packages: @lssm-tech/lib.ai-agent, @lssm-tech/integration.voice-input-web, @lssm-tech/integration.voice-input-native

    1. Import graphOpsTools from src/tools/graph-ops/graphOps.command.ts.
    2. Register them via the ai-agent tool registry.
    3. Wire AiSuggestionConfirm to gate all tool outputs behind explicit user confirmation.
    4. Set per-session budget ($0.50) and per-user/day budget ($5.00) in useAiCostGuard.

  • Wire createWebVoiceInput for web STT

    manual

    Use voice-input-web for Web Speech API primary + Deepgram fallback on web surfaces.

    Packages: @lssm-tech/lib.ai-agent, @lssm-tech/integration.voice-input-web, @lssm-tech/integration.voice-input-native

    1. Install @lssm-tech/integration.voice-input-web.
    2. Call createWebVoiceInput({ deepgramApiKey, language }) with the key from env vars only.
    3. Wire onTranscript to the AI input slot; wire onError to the error surface.
    4. Do not import this package in native (.native.tsx) files.

  • Wire createNativeVoiceInput for native STT

    manual

    Use voice-input-native for on-device iOS STT and Deepgram fallback on Android.

    Packages: @lssm-tech/lib.ai-agent, @lssm-tech/integration.voice-input-web, @lssm-tech/integration.voice-input-native

    1. Install @lssm-tech/integration.voice-input-native.
    2. Add @react-native-voice/voice and expo-av to native build deps.
    3. Run npx expo prebuild after adding this package.
    4. Call createNativeVoiceInput({ deepgramApiKey, language, preferOnDevice }).
    5. Do not import this package in web files.

Unique release changes

  • - Add graph.search, graph.filter, graph.cluster, and graph.suggest to the agent tool registry for graph surfaces.

    3 packages · 3 occurrences

  • - M3 — Four graph AI LLM tools (graph.search, graph.filter, graph.cluster, graph.suggest) registered in ai-agent tool registry; two new voice input integration packages (voice-input-web, voice-input-native).

    3 packages · 3 occurrences

  • - Use voice-input-native for on-device iOS STT and Deepgram fallback on Android.

    3 packages · 3 occurrences

  • - Use voice-input-web for Web Speech API primary + Deepgram fallback on web surfaces.

    3 packages · 3 occurrences

Impacted packages

  • @lssm-tech/integration.voice-input-native

    Layer: integrations · 4 changes

  • @lssm-tech/integration.voice-input-web

    Layer: integrations · 4 changes

  • @lssm-tech/lib.ai-agent

    Layer: libs · 4 changes