AI SDK Patterns

Custom Agent implementation demonstrating the Agent interface abstraction with an orchestrator that routes queries to specialized sub-agents (research, analysis, support). Shows options passing and structured outputs.

Route user queries to specialized AI agents based on context and intent. Includes dynamic agent selection, load balancing, and fallback handling.

Give an Eve agent tenant-scoped long-term memory with dynamic instructions and three ordinary tools — remember, list, forget — with scope derived from verified session auth. Runs without provider credentials.

Cron-driven Eve agent that gathers metrics on a schedule, delivers proactive digests, and lets tenants manage their own dynamic reminders. Runs without provider credentials.

Build and verify a filesystem-first Eve agent with a typed weather tool and deterministic eval suite. Runs without provider credentials.

Multi-agent support triage on Eve: a coordinator classifies requests and delegates to declared billing, technical, and escalation specialists — each with its own tools and instructions. Runs without provider credentials.

Dynamically configure one ToolLoopAgent per request with callOptionsSchema and prepareCall. A customer-support agent rewrites instructions, gates tools, tunes reasoning effort, and injects RAG context based on account tier and urgency.

Inject request-scoped context (user identity, database handles, session) into ToolLoopAgent tools via experimental_context, so the model never supplies sensitive identity in tool inputs.

Demonstrate ToolLoopAgent loop control with stopWhen. Covers stepCountIs, hasToolCall, combined conditions, and a custom StopCondition with step timeline visualization.

Demonstrate ToolLoopAgent prepareStep for trimming long message history. Pre-seeded 24-message conversation shows context compression on the first run with per-step trim telemetry.

Lightweight ToolLoopAgent pattern for Anthropic ephemeral cache control through Vercel AI Gateway: cached system instructions with cacheControl and reports cache read/write tokens.

Lightweight ToolLoopAgent pattern for OpenAI prompt caching through Vercel AI Gateway: stable system instructions, promptCacheKey, and per-turn cachedPromptTokens metrics.

Minimal ToolLoopAgent chat demo: on-call-style skills (incident triage + stakeholder comms) on disk, skill tool loads SKILL.md + paths, readFile pulls reference checklists and templates from the sandbox.

Demonstrate structured agent outputs using ToolLoopAgent with Output.array. Returns structured array output with element schema for predictable, type-safe responses.

Demonstrate structured agent outputs using ToolLoopAgent with Output.choice. Constrain agent outputs to predefined choices for predictable, type-safe responses.

Process content with multiple AI agents running simultaneously. Demonstrates concurrent analysis for faster results.

Improve output quality with dual agents. Evaluator analyzes quality while optimizer refines results through iterative feedback.

Execute multi-step workflows with typed tools. Includes streaming, tool chaining, and decision-making for automated tasks.

Coordinate multiple worker agents for project management. Handles task distribution, progress tracking, and result synthesis.