Pricing and Limits

eve usage is billed through the Vercel resources and third-party services your agent uses. An eve deployment can use Vercel Functions for compute, Vercel Workflows for durable sessions, Vercel Sandbox for isolated command execution, and AI Gateway or model providers for model calls.

• Session volume: Each session and turn invokes agent routes and creates Workflow events.
• Model usage: Prompt length, tool results, reasoning, cached tokens, and output length affect model costs.
• Tool calls: External API calls can increase Function duration, Workflow event count, and third-party usage.
• Streaming: Stream writes are persisted by Workflows and count toward Workflow data written and retained.
• Sandbox usage: Command execution, provisioned resources, network transfer, snapshot storage, and sandbox creation can add Vercel Sandbox usage.

eve inherits limits from the platform products behind the agent:

• Function limits: Agent route runtime, memory, bundle size, request body size, concurrency, and regional behavior follow Vercel Functions limits.
• Workflow limits: Session durability, event creation, stream chunks, hook tokens, replay duration, payload size, and run storage follow Workflow run limits.
• Sandbox limits: Sandbox runtime, vCPU, memory, network, snapshot, and tag limits follow Sandbox pricing and limits.
• Model limits: Context windows, rate limits, output limits, and provider availability depend on the selected model and provider.
• Channel limits: Slack, HTTP clients, and other channel platforms can impose their own request, payload, retry, and rate limits.

To estimate production usage, start with one representative agent task and record:

1. How many turns the session takes.
2. How many model calls and tool calls each turn makes.
3. The input, cached, and output tokens per model call.
4. The amount of data streamed to the client.
5. Whether the task uses sandbox execution, snapshots, network access, or long-lived persistence.

Then map those measurements to the resources above. Agent Runs shows token counts, timing, tool calls, and turn breakdowns, making it a practical starting point for real usage estimates.

• Use smaller or cheaper models for low-risk turns, and reserve larger models for tasks that need them.
• Keep tool results and message history concise so the model receives only relevant context.
• Move repeatable instructions into skills so they load only when relevant.
• Use sandbox network policies and lifecycle hooks to limit unnecessary network and compute usage.
• Break very large or long-running jobs into smaller sessions or subagents when they approach Workflow replay or storage limits.
• Configure Spend Management and project alerts for production agents.

• Concepts: learn how eve's runtime pieces fit together.
• Agent Runs: inspect agent sessions and configure trace collection.
• Observability: configure OpenTelemetry exports and policies.