Langfuse roadmap

Langfuse is open source, and we want to be transparent about where the project is going. This page summarizes the product areas we are currently investing in.

Each quarter, we host a Langfuse town hall to demo new features and discuss what's next. Watch the latest recording below or subscribe to our event calendar for updates on the next one.

🚧 In progress and coming next

New capabilities

• General availability of the AI Gateway
  • Broaden model and provider support.
  • Add fallbacks to keep requests flowing when a model or provider is unavailable.
  • Add spend caps to control costs per gateway key.
  • Provision users via SCIM.
  • Add guardrails to enforce policies on requests and responses.
• Proactive issue detection
  • Develop Topics to identify recurring user intents, behaviors, and failure patterns across traces at scale.
  • Surface clusters, outliers, and representative examples so teams can find issues without reviewing every trace.
• Agent transcripts
  • Extend a shared transcript representation across traces and sessions to make long-running agent interactions easier to read.
  • Use transcripts as a foundation for Topics and other improvement workflows.
• Halo improvements
  • Help users complete workflows such as LLM-as-a-judge calibration with more capable sandbox tools.
  • Introduce Halo loops to automate recurring workflows.
  • Introduce Halo reports and apps to build custom views on top of your Langfuse data.
• MCP server improvements
  • Add OAuth authentication to the MCP server so it can respect user-level permissions.
  • Improve reliability and expose more opinionated tools to help agents perform better.
• Agent view
  • Provide one overview page for each agent observed with Langfuse.
• Evaluation
  • Evaluate complete agent interactions with multi-span, session, and trajectory evaluations, using the observations-first data model.
  • Add native support for the Harbor framework to run evaluations and experiments in sandboxed environments.
  • Add semantic value ranges to evaluation scores so results are easier to interpret.
• Security and authentication improvements
  • Add granular permissions to API keys, including gateway-specific access.
  • Support user-owned and service-account API keys.
  • Add organization-level API keys that can write across multiple projects.
  • Support custom user roles and user groups.
  • Restrict an organization to your email domain and automatically invite new teammates.
  • Add read-level audit logs that record who viewed which data, not only who changed it.
• Usage visibility and cost controls
  • Break down Langfuse usage by project, environment, and span name for organization admins.
  • Control ingestion with server-side filters, sampling, usage limits, and data masking (including PII) before data is stored.
• Data platform
  • Expose the query builder via API and MCP to run flexible metrics queries against your Langfuse data, programmatically or from coding agents.
  • Speed up metadata analytics with performant group-bys and aggregations.
• UI/UX
  • Continued improvements to the new unified UI.
  • UI accessibility.

Self-hosting improvements

• Batch more S3 ingestion events to reduce cost.
• Move ingestion work to Rust in stages to reduce latency and cost.
• Make sure all new features are available self-hosted and across cloud providers.

✅ Recently released

The 10 most recent changelog updates:

• Microsoft Azure and Google AI Studio in the AI Gateway(Nov 12, 2026)
• Route AI Gateway connections through your own endpoint(Nov 11, 2026)
• Call decision models through the AI Gateway(Nov 11, 2026)
• Create embeddings through the AI Gateway(Nov 11, 2026)
• Set the default evaluation model and run batch evaluations via MCP(Nov 11, 2026)
• Call Amazon Bedrock models through the AI Gateway(Nov 10, 2026)
• Call OpenRouter, Vercel AI Gateway, and xAI models through the AI Gateway(Nov 10, 2026)
• Run code evaluators on your own executor(Nov 10, 2026)
• See when token usage is missing(Nov 6, 2026)
• Connect your S3 buckets with an IAM role(Nov 6, 2026)

Shape the roadmap

Your feedback helps us decide what to prioritize:

• Suggest a feature or improvement, or vote on ideas from other users.
• Report a bug.
• Discuss your use case with the community on Discord.
• Join a community hour to talk with the Langfuse team.

---

---