Vercel Agent public beta: chat, investigations, and approved actions

Vercel has announced the public beta launch of an expanded version of Vercel Agent, its artificial intelligence assistant integrated into the deployment platform. This update transforms the agent from a simple code reviewer into a full-fledged assistant that lives in the dashboard and can interact with real project data. Below, we analyze in detail what has changed, why it matters, and what implications it has for the development ecosystem.

What happened?

The original Vercel Agent, launched in 2024, focused on suggestions in pull requests via a sandbox. Now, in public beta, the agent expands its capabilities to three key areas: contextual chat, production investigations, and approved actions. According to Vercel's official blog, the agent "now lives in your dashboard and can investigate production issues, answer questions about your projects, and act on your behalf." This is possible because the agent has access to internal platform signals: deployments, logs, metrics, project configuration, usage, and connected repositories. That context is what turns a question into an answer and a problem into a solution.

• Contextual chat: Users can ask questions about their projects and get answers grounded in real deployment data, directly from the dashboard. For example, querying why a deployment failed or how a domain is configured.
• Production investigations: The agent can analyze deployment failures, runtime errors, cost spikes, or performance issues, tracing the root cause and recommending solutions. This includes the ability to generate corrective code.
• Approved actions: With explicit user authorization, the agent can open pull requests, revert deployments, modify configurations, or execute other actions requiring elevated permissions. Generated code is first run in a Vercel sandbox before reaching production.

The agent operates under its own identity and respects the requesting user's permissions. It is read-only by default; any action requiring elevated permissions must be explicitly approved. Additionally, actions are attributed to the agent, the requester, and the approver, ensuring traceability.

Why is this important?

This evolution marks a significant step toward assisted automation in the development lifecycle. Unlike generic assistants like GitHub Copilot or ChatGPT, which lack access to deployment context, Vercel Agent has full visibility into the ecosystem: logs, metrics, configurations, and connected repositories. This allows responses to be grounded in real data, not general knowledge. As Vercel notes, "that context is what turns a question into an answer and a problem into a solution."

Furthermore, the control model addresses a key concern in AI adoption: governance. By requiring explicit approval for elevated actions and attributing each action to multiple entities, Vercel offers traceability and security. This contrasts with more lax approaches from other AI tools that may make changes without clear oversight. The ability to run code in a sandbox before production adds an extra layer of safety.

The timing is also relevant: in 2024, Vercel launched the Vercel AI SDK and developer tools, but this is the first time they integrate a conversational agent directly into the production dashboard. This positions Vercel as a leader in integrating AI into the development workflow, competing with platforms like Netlify (which has Netlify AI) and AWS (with Amazon Q Developer).

Consequences and outlook

For development teams, this can significantly reduce the time to diagnose production incidents. Instead of manually reviewing logs or using multiple tools, developers can ask the agent directly. For example, an unexpected cost spike can be automatically investigated, with the agent identifying the cause (such as an over-executing serverless function) and suggesting an optimization. The ability to open corrective PRs further accelerates resolution.

However, reliance on an agent with access to sensitive data requires careful permission configuration. Organizations with strict compliance requirements (such as SOC 2 or HIPAA) will need to assess whether the current permission model is sufficient. Vercel states that the agent respects user permissions, but access to logs and metrics could expose sensitive information if not configured correctly.

The pricing model combines a limited free period for simple queries (like finding a setting or explaining a DNS warning) with a per-token cost for investigations and actions: $0.25 per million Vercel tokens, plus the underlying provider cost at no markup. This can be competitive against solutions like AWS Assistant (which charges per query) or GitHub Copilot (which has a fixed per-user price). During the beta, simple queries are free for a limited number of messages, allowing teams to test functionality without upfront cost.

“Simple requests, like finding a setting, explaining a DNS warning, or answering a docs question, are free for a limited number of messages during the beta.” — Vercel Blog

The beta is available for Pro and Enterprise teams, with gradual rollout. Users can request access via a form. This suggests Vercel is testing scalability and gathering feedback before a general release.

What should readers know?

• The agent is read-only by default; actions require explicit user approval.
• Investigations and actions are billed per token consumed, with a cost of $0.25 per million Vercel tokens plus the underlying AI provider cost (no markup).
• The agent already includes the existing code review functionality, now as one capability within the expanded agent.
• Full documentation is available at vercel.com/docs/agent.
• The agent is available in public beta only for Pro and Enterprise teams, with gradual access. Hobby plan users do not have access yet.
• Elevated actions (like opening PRs or reverting deployments) are attributed to the agent, the requester, and the approver, ensuring traceability.

Compared to previous events, such as the launch of the Vercel AI SDK in 2024, this update represents a qualitative leap: not only is an AI API offered, but an integrated agent that acts within the platform. This could mark the beginning of a trend where deployment platforms compete to offer intelligent assistants that automate operational tasks. However, adoption will depend on developers' trust in delegating critical actions to an AI agent. The public beta will allow Vercel to refine the product based on real usage before a general release.