What's new in LangWatch
Every release, improvement, and fix as it ships. Follow along as the platform evolves.
Instant Evals in the Trace Explorer, and Files in Datasets
Ask the Trace Explorer a question no filter can answer, and upload images and files into dataset cells.
Instant Evals: Search Your Entire Production History
Ask a question of every conversation you have stored, and get the matches back in minutes.
Go SDK 1.0 and Search in Datasets
Go SDK 1.0, search in the dataset editor, and agents discovered from connected providers.
Agent Testing v2 and Prompt Optimization in the Workbench
Agent Testing replaces Simulations, comparison runs arrive, and Langy optimizes prompts in the workbench.
New Navigation and Voice APIs support on the Gateway
The sidebar becomes a product switcher and an icon rail, and realtime voice bills to a virtual key.
A Governed SQL Workbench, and the Go SDK
Query your data in a governed SQL workbench, and instrument Go.
Coding Agent Capture, and Gateway Billing
Capture Copilot and Claude Code sessions, and bill gateway spend with webhooks and reconciliation.
The AI Gateway Grows Budgets
Budgets on every dimension of gateway traffic, with live spend and a redesigned key drawer.
Audio in Traces, and a Terminal Tab
Compare two variants without a reference answer, and play audio inside a trace.
Cost Intelligence and Pairwise Compare
Spend breaks down by category, and experiments compare two variants head to head.
Datasets, Rebuilt on S3
Datasets store in S3 as chunked JSONL, with CSV and bulk upload.
Studio & Datasets Makeover, Scoped Data Privacy, and TTFT
Datasets edit in place, privacy rules redact secrets and PII, and TTFT becomes a column.
Traces v2 Leveled Up, and Model-Cost Preview
See which spans a cost rule matches before you save it, plus two rounds of trace explorer work.
Langy for Everyone, and Voice Agent Testing
Langy opens for everyone on a project, and Scenario tests voice agents end to end.
The AI Governance Platform
One dashboard for spend, usage, anomalies and policy across every AI tool your teams run.
Keyboard Warriors - The Command Bar
A Cmd+K command bar to navigate anywhere, search for any entity, switch themes, and discover a few surprises along the way.
Polished Analytics
A sharper, more polished analytics experience - sticky headers, better empty states, dark mode refinements, and fixes for filter behavior across the board.
LangWatch 3.0 - Self-Host with ClickHouse
LangWatch 3.0 is here, a simpler, faster stack built on ClickHouse. Deploy with Helm, run with Docker Compose, no Elasticsearch required.
Event Sourcing Engine Fast is the only “mode”
LangWatch now runs on a purpose-built event sourcing engine. Traces, evaluations, and simulations are processed in real-time with no accuracy trade-offs
Built-in MCP Server with OAuth
LangWatch now ships a built-in MCP server with OAuth. Connect Claude Code, or any MCP client to manage prompts, datasets, evaluators etc from your editor.
Prompt Tags
Tag your prompts across every surface platform, SDKs, CLI, and MCP. Use syntax (prompt:tag) to fetch tagged versions, and manage tags from wherever you work.
LangWatch Scenario on the Platform
Create and run agent simulation tests directly in LangWatch, no code required. Define user personas, set pass/fail criteria, and run scenarios against HTTP.
Graph Alerts
The alerting feature now includes custom graph alerts that trigger when specific thresholds are met, with notifications sent via preferred channels like Slack.
Audit Logs
We’ve introduced a new Audit Logs page, giving you complete visibility into user activity across your projects. This feature helps you monitor actions...
Better Agents just got better! Support for Google ADK, LangGraph, Antigravity & more
New Coding Assistants Supported: Antigravity, Crush CLI, Gemini CLI, Qwen Code. With this, Better Agents now works across a much wider range of tools and...
Prompts: Fetch Policy
Choose how a prompt is fetched: cached, or fresh on every call.
Replicate Workflows/Prompts/Datasets
We’ve added the ability to replicate workflows, their datasets, and prompts to another project. This lets you safely test in a staging environment, validate...
Launch Week - Day 5: Better Agents
Better Agents is a CLI tool and a set of standards for agent building, it supercharges your coding assistant...
Launch Week - Day 4: Writing Tests with the Scenario MCP
Scenario MCP will enable you to generate high-quality tests, and it's as easy as cake. One prompt is all it takes, and the outcome will...
Launch Week - Day 2: RBAC Custom Roles & Permissions
With the new RBAC Custom Roles, you can now enforce finer-grained access control, allowing different role sets to focus on their specific domains. The permis...
Launch Week - Day 3: Testing Voice Agents with LangWatch Scenario in Real Time
LangWatch Scenario now supports testing voice-powered AI agents using OpenAI's Realtime API through a new RealtimeAgentAdapter. This enables comprehensive...
Launch Week - Day 1: LangWatch New Prompt Playground
With the new RBAC Custom Roles, you can now enforce finer-grained access control, allowing different role sets to focus on their specific domains. The permis...
Add Member, Reworked
Invite someone and set their org and team roles in one dialog.
Thread-level Evaluations
Score a whole conversation instead of a single message.
Real-time simulation visualization and debug mode
Watch simulated conversations play out in real-time with parallel execution, versioned runs, and interactive debugging.
Scripted simulations, evaluations, and guardrails
Scripted simulations let you orchestrate how conversations unfold, when evaluations occur, and what custom logic runs.
Test agents on Mastra, Agno, and 10+ other frameworks
Scenario is framework-agnostic, supporting any agent architecture through the AgentAdapter interface.
Introducing simulation-based agent testing
Test your agents with scenarios directly in your codebase using Python and TypeScript.