Atlas
Product · SaaS / cloud software · Engineering & development by Voiceflow
Your AI copilot for building better agents.
Atlas is an AI copilot for Voiceflow that analyzes customer transcripts to identify failures and propose agent improvements. It allows teams to build, test, and refine AI agents using plain language while maintaining human-in-the-loop control.
- Launched
- 2026-08-13
- Languages
- English
Category: Conversational AI
- Platforms
- Web-based workspace
- Availability
- Beta (as of August 2026)
- Latency
- 50ms Voiceflow platform latency
- Scale
- 300,000 messages per minute
- Modes
- Safe mode and Autonomous mode
- Developer Access
- API, CLI, and MCP server
- AI copilot for building, testing, and improving agents
- Safe mode for human-in-the-loop approval of changes
- Automated regression test generation from transcripts
- Scheduled weekly reviews of conversation data
Overview
Atlas is Voiceflow's AI copilot designed to help enterprise teams find problems, make changes, and test improvements within their AI agent's workspace. It serves as a "co-captain" for support, CX, and engineering teams by reading every conversation to identify where an agent failed and proposing specific fixes for human approval. Users can interact with Atlas in plain language to generate new playbooks, fill knowledge gaps, and create regression tests from real customer evidence. The platform offers a Safe mode for read-only investigation and an Autonomous mode for teams that want to grant the copilot more direct action. By turning raw conversation data into actionable improvements, Atlas aims to reduce the manual effort required to maintain high-performing AI agents.
Investigation and Analysis
- Automated Conversation Review: Atlas reads every conversation to find where the agent failed and brings the specific fix to the user for approval.
- Natural Language Building: Users can ask Atlas in plain language to build, test, and improve the agent across the entire application.
- Insight to Outcome: Atlas turns conversation evidence into a focused improvement plan by identifying where users get stuck.
- Recurring Problem Identification: The copilot identifies repeat requests worth fixing before any changes are made to the agent.
Building and Testing
- Direct Agent Edits: Atlas helps edit the agent directly where it lives, such as clarifying a policy or updating prompt instructions.
- Regression Testing: Atlas turns real conversations into regression tests to ensure that previously fixed issues do not return.
- Verification Workspace: Users can verify the results of changes and improvements within the same workspace.
Control and Operations
- Safe Mode Governance: In Safe mode, Atlas performs read-only investigations while project changes are paused for human approval.
- Autonomous mode is available for teams that choose to allow the copilot more direct action.
- Activity Inspection: Atlas shows its activity and the tools it calls, allowing users to inspect proposed edits before implementation.
- Scheduled Routines: The copilot runs on a schedule, such as performing a weekly review of the previous week's conversations.
Use cases
- Support teams use Atlas to identify recurring customer problems and generate new playbook instructions to resolve them.
- Conversation designers ask Atlas in plain language to create regression tests based on real-world failure cases.
- CX leaders use Atlas to get concise explanations of trends in conversation volume and performance metrics.
- Engineers connect Atlas to their development workflow via MCP to automate agent testing and improvement.
Specifications
- Deployment
- Cloud-based platform with private cloud hosting available on Enterprise plans
- Platforms
- Web-based workspace
- Integrations
- Zendesk, Salesforce, HubSpot, Shopify, Airtable, Google Sheets, Gmail, Twilio
- APIs and SDKs
- Voiceflow APIs and MCP server expose the same underlying capabilities used by Atlas
- Sign-in and administration
- SSO, five workspace roles, and project-level permissions available on Enterprise plans
- Languages
- Supports multilingual agent development including Turo's multilingual support chatbot
- Availability
- Beta (as of August 2026)
- Latency
- 50ms Voiceflow platform latency
- Scale
- 300,000 messages per minute
- Modes
- Safe mode and Autonomous mode
- Developer Access
- API, CLI, and MCP server
How it works
- Analyze Transcripts: Atlas reads every conversation transcript to identify where the agent failed or where users became frustrated.
- Review Recommendations: The copilot generates a focused recommendation, such as a playbook change or a new test, based on the conversation evidence.
- Approve and Apply: Users approve the proposed edits or tests, which Atlas then applies to the agent within the workspace.
- Verify Improvements: Atlas runs simulations and regression tests to verify that the changes improved the agent's performance.
Added by AntSpace from public sources (voiceflow.com).