← Back
ElevenLabs ships agent updates with OAuth auth, conversation search, and file uploads
· releaseapifeaturesecuritysdk · elevenlabs.io ↗

ElevenAgents Updates

ElevenLabs introduced significant enhancements to its agent platform, focusing on security, conversation management, and authentication capabilities.

Security & Authentication

MCP servers now support advanced authentication methods including OAuth2 Client Credentials, Basic Auth, Bearer Auth, JWT, and custom header authentication. Developers can select auth connections when creating or editing MCP servers, with automatic token refresh and authentication header handling. This enables secure, production-grade integrations with third-party services.

Conversation Management

Two new search endpoints allow developers to query conversation data:

  • Text search: Full-text and fuzzy search over conversation transcripts
  • Smart search: Semantic search using embeddings for more intelligent queries

Additionally, file upload and deletion endpoints now let agents work with documents and files within conversation contexts.

Privacy & Safety

A new conversation history redaction feature lets developers configure entity redaction (names, email addresses, etc.) before transcripts and analysis are stored. Two new guardrail types were added:

  • focus: Keeps conversations on-topic
  • prompt_injection: Detects and prevents prompt injection attempts

The legacy alignment guardrail was removed.

Other Improvements

  • New LLM information endpoint for programmatic discovery of available models, deprecation status, and capabilities
  • Support for qwen3_embedding_4b embedding model in RAG indexing
  • New workspace-level permissions for image and video generation
  • Music generation now supports workspace sharing

SDK & API Updates

JavaScript SDK (v2.36.0): Improved Speech-to-Text type safety and added support for all new API features.

Python SDK (v2.36.1): Added missing music generation parameters (seed, loudness, quality, guidance_scale).

Removed endpoints: Legacy voice generation endpoints (/v1/voice-generation/*) were deprecated in favor of Voice Design and Voice Remix.

Updated MCP endpoints: Now support auth_connection field and new input_overrides for tool parameter customization.