Vercel Eve Agent Observability Now Built Into the Dashboard
Quick Reads
- Vercel’s eve agent framework now has a built-in Agent Runs tab in the project dashboard
- Each session shows trigger, duration, and token usage at a glance
- Developer and Business mode serve engineers and non-technical reviewers separately
- Run data is encrypted; retention spans 12 hours to 3 days depending on the plan
- African dev teams building autonomous agents gain powerful debugging without extra tooling
Vercel has shipped Agent Runs, a built-in observability view for its eve open-source agent framework, accessible directly from the Vercel dashboard. Teams therefore get full session tracing with no OpenTelemetry setup required. This is a meaningful step for teams shipping agents to production.
Until now, debugging agent failures required digging through raw function logs. That was tedious and error-prone. The Agent Runs tab now appears automatically for every eve project. It surfaces trigger, duration, and token usage for each session at a glance. Developers can furthermore drill into any run to inspect every turn, model call, and tool call. Runtime errors that previously vanished into function logs now correlate directly to the failing step.
Vercel also introduced two distinct viewing modes. Developer mode shows raw tool names, input and output JSON, and per-step token counts. Business mode, however, humanises tool names, hides JSON, and generates a plain-English summary so a non-technical reviewer can audit what an agent did. This dual-mode design acknowledges that Vercel eve agent observability isn’t just a developer concern, product leads and compliance teams need it too.
On data retention, run data is encrypted by default. Retention scales with plan: 12 hours for Hobby, 1 day for Pro, and 3 days for Enterprise. Additionally, Pro and Enterprise teams can extend retention to 30 days with Observability Plus.
For teams that already use a custom OpenTelemetry backend, AI SDK spans can still be exported from agent/instrumentation.ts to destinations like Braintrust or Datadog.
The eve framework itself launched in June with a filesystem-first approach to building agents. An agent is simply a directory of files. eve runs it with durable execution, a sandbox, approvals, and evals built in. Consequently, it works with any model or MCP server. Vercel reports running more than 100 agents on eve in production today, including a data analyst agent handling 30,000 questions monthly.
For developer teams across Nigeria and Africa building autonomous workflows, Vercel eve agent observability removes the need to stand up a separate monitoring stack. That significantly reduces operational burden, especially for lean engineering teams shipping AI-powered products.
To get started, visit the eve observability documentation or open Agent Runs directly in your Vercel project dashboard.
