Enterprise Eve Agent Development
A practical enterprise handbook for building high-quality durable AI agents with Vercel Eve, based on Eve 0.11.4 source and the full official docs corpus.
Recently Updated Chapters
Interpret defineAgent configuration as runtime policy for models, output schemas, compaction, and experimental flags.
How Eve channels own session creation, continuation tokens, route auth, NDJSON streams, and frontend/client integration.
How Eve transforms source manifests into compiled manifests and runtime agent graphs.
Design context, skills, dynamic tools, dynamic instructions, and dynamic skills for high-quality Eve agents.
Combine Eve features into support, research, code, back-office, analytics, and incident-response agent patterns.
Eve is Vercel's filesystem-first framework for durable backend AI agents. Instead of hiding an agent inside one large configuration object, Eve turns files under agent/ into a discovered, compiled, executable, durable runtime graph.
This handbook is not a beginner tutorial. It is written for engineers who already build, operate, or govern AI agents: platform engineers, AgentOps owners, staff engineers, and teams running internal automation or coding agents. The focus is authority boundaries, durability, evaluation, observability, approval, sandbox isolation, multi-agent design, and production governance.
Analysis Baseline
The source baseline is a private local Eve checkout, with packages/eve/package.json
reporting Eve 0.11.4. On 2026-06-18, this handbook was also checked against
Eve official docs and all 67 /docs/* pages listed in the
official sitemap.
Beta Baseline
The official site presents Eve as Beta. Enterprise adoption should treat API stability, Vercel beta terms, provider terms, telemetry exports, and approval policies as explicit release gates.
Why Eve Matters For Enterprise Agents
Eve's value is not only defineTool(). The deeper value is the way it separates authoring, runtime, durability, and delivery boundaries.
| Boundary | Eve design | Enterprise meaning |
|---|---|---|
| Authoring | agent/ file slots | Reviewable agent capability surface |
| Execution | app runtime vs sandbox | Separates secrets from model-driven shell/file work |
| Durability | session -> turn -> step | Long-running work survives restarts and redeploys |
| Operations | hooks, evals, OTel, Agent Runs | Quality, cost, and security become review loops |
Core Architecture
Operating Principles
| Principle | Practical standard |
|---|---|
| File slots are APIs | Manage tools/, skills/, connections/, and subagents/ as reviewable authority units. |
| The model sees minimum necessary context | Use toModelOutput, skills, dynamic context, and sandbox files to avoid prompt overload. |
| Side effects must survive replay | Use idempotency keys, approval gates, and external ledgers for non-idempotent writes. |
| Channels own session identity | continuationToken is not a queue; route auth and session ownership are separate controls. |
| Quality is a release gate | Use eve eval --strict, audit hooks, OTel, and $eve.* workflow tags as deploy criteria. |
Contents
Part 1. Core Architecture
Ch1. Eve Mental Model
Understand Eve through filesystem authoring, durability, channels, and the harness.
Ch2. Source Code Atlas
Read packages/eve/src by implementation responsibility.
Ch3. Project Layout and Discovery
See how agent/ slots become a source manifest.
Part 2. Agent Quality Design
Ch5. agent.ts, Models, Compaction
Set model routing, output schemas, compaction, and experimental flags as operational policy.
Ch6. Context, Skills, Dynamic Capabilities
Design context surfaces for high-quality, multi-tenant agents.
Ch7. Tools, Approval, Connections
Combine authored tools, OAuth, MCP/OpenAPI connections, and human approval safely.
Part 3. Operational Runtime
Ch9. Channels, Auth, Streaming
Design route auth, continuation tokens, NDJSON streaming, and custom channels.
Ch10. Subagents, Workflows, Remote Agents
Use local subagents, remote agents, and experimental workflow orchestration.
Ch11. Schedules, State, Hooks
Connect cron, durable state, and stream-event hooks to operations.
Part 4. Production Operations
Ch13. Observability and Deployment
Use OTel, workflow tags, and Vercel/self-host deployment checklists.
Ch14. Enterprise Patterns
Apply Eve to support, research, coding, back-office, analytics, and incident response agents.
Ch15. Migration and Governance
Migrate existing agents and establish operating governance.