Edge

Edge deployment is useful when the agent entry point needs to sit near users, devices, or regional data sources.

Good edge candidates

Use edge deployment for:

• request routing
• lightweight policy checks
• identity and ACT verification
• cached model or tool responses
• WASI-compatible agent steps
• low-latency tool dispatch

Avoid edge deployment for heavy durable workflows, large vector stores, or provider calls that require region-specific secret handling your edge platform cannot support.

Required boundaries

Edge agents should keep:

• strict network egress rules
• short execution budgets
• explicit memory limits
• deterministic fallback behavior
• clear telemetry export
• no raw production seeds in edge bundles

Relationship to WASM

WASM and WASI are the preferred packaging direction for edge-compatible agent logic because they make host capabilities explicit.

See WASM and WASI.