Integration Boundary Guide
Agent Integration Boundary Guide
Use this guide when deciding how Gait fits with your agent runtime, especially managed/preloaded agents.
Core Rule
Fail-closed enforcement requires an interception point before real tool execution.
If you cannot intercept tool calls, Gait can still add observe/report/regress value, but it cannot block side effects inline.
Integration Tiers
Tier A: Full Runtime Interception (Best Fit)
Examples:
- wrapper/decorator in your app runtime
- local sidecar calling
gait gate eval - local
gait mcp servewith caller-side enforcement
What Gait can do:
- full policy enforcement (
allowonly executes) - approval/delegation token checks
- signed trace emission
- runpack/pack capture and regress loop
What remains external:
- model selection/planning/orchestration logic
Tier B: Middleware/API Boundary Interception
Examples:
- API gateway/middleware for tool execution endpoint
- service boundary where tool payloads can be normalized and evaluated
What Gait can do:
- enforce decisions at middleware boundary
- emit signed traces and artifacts
- provide deterministic CI regression loop
Constraints:
- normalization quality depends on payload completeness at boundary
- enforcement reliability depends on strict middleware placement
Tier C: Managed/Preloaded Agent Products (Limited Interception)
Examples:
- hosted copilots/preloaded agents where tool execution path is not user-controlled
What Gait can do:
- observe/report if traces/artifacts can be exported
- post-hoc verification and regress generation from exported evidence
- policy simulation and CI policy tests with representative fixtures
What Gait cannot do (without interception):
- guaranteed inline block before side effects
- strict runtime fail-closed enforcement
Decision Flow
- Can you intercept every state-changing tool call before execution?
- If yes: use Tier A and enforce non-
allowas non-executable. - If partially: use Tier B and close bypass paths first.
- If no: use Tier C and treat controls as observe/report/regress until interception is available.
Practical Paths
- Tier A quickstart:
examples/integrations/openai_agents/quickstart.py - Tier A/B transport path:
gait mcp serve - Tier C fallback:
gait report top,gait regress init,gait regress run
Related Docs
docs/flows.mddocs/integration_checklist.mddocs/mcp_capability_matrix.md