Where PlanVault fits
This page explains how PlanVault sits alongside common stack layers—without claiming to replace specific vendor products or open-source libraries. The focus is the governed API and event execution boundary.
Where PlanVault sits next to your agent stack
PlanVault is the governance and execution layer that connects to your agent stack — existing LangChain, LangGraph, CrewAI, AutoGen, DSPy, OpenAI agent API, or MCP agents can run behind a PlanVault tool boundary, with tool calls, secrets, approvals, and recovery handled outside the model and every run recorded for audit and replay.
PlanVault keeps tool calls, secrets, approvals, and recovery outside the model.
Tool routing
Centroid-based DB routing; shortlist built before the LLM
Managed by the relevant OpenAI API surface
Often implemented in application logic or through LLM/graph routing
Often policy, catalog, or app-level routing
Self-hosted · VPC-ready
External API dependency
Self-managed infra, cloud LLM
Often SaaS-oriented; options vary by vendor
Catalog scalability
Dynamic routing from catalogs of thousands of tools; adaptive shortlist per step
128 tools (Assistants API); context-bounded (Responses API)
Depends on agent design and available context
Often manually or policy-managed catalog
Crash recovery
State and recovery depend on the provider-managed API surface
Checkpointing is possible; recovery semantics often stay app-owned
Often tied to the vendor conversation or workflow model
Large responses
Schema flattening, JSONPath extraction, stdlib tools, depth truncation
Depends on token/context limits and API pattern
Usually handled at the application layer
Depends on product focus and integrations
Categories next to PlanVault
Workflow runtimes
Workflow engines excel at human-oriented processes and SLAs; PlanVault adds a layer for AI-initiated actions: bounded tool catalogs, execution policy, and a complete event trail.
Agent orchestration frameworks
Libraries and frameworks help describe agent graphs and logic; PlanVault focuses on what happens after planning: validation, FSM execution, journaled transitions, and side-effect policy.
Enterprise governance systems
Compliance and IAM platforms set access rules for people and services; PlanVault adds a boundary for AI runs: plan validation, tool execution, and reviewable evidence.
Tool routing and retrieval middleware
Retrieval and routing narrow context for the model; PlanVault holds the execution boundary: who may call which API, with which secrets, and with what audit trail after the answer.
Durable execution systems
Systems that guarantee long-running steps often do not define per-AI-run API call policy, human approvals, and audit evidence—that is PlanVault’s boundary.
AI runtime safety controls
Moderation and guardrails reduce unsafe model outputs; PlanVault governs what happens in your systems: calls, approvals, idempotency, and recovery.
Contrast: planning versus governed execution
- An LLM or agent produces or proposes a plan of actions.
- PlanVault validates the plan and executes through runtime and FSM—rather than trusting unconstrained API calls.
- State transitions are journaled; side effects are policy-gated.
- Recovery, audit, replay, cost visibility, retention, and user-data operations are runtime-boundary concerns—not only model text.
Limitations and transparency
PlanVault is designed for customer-controlled deployment, network integration, automated SBOM publication, security review, and penetration-test readiness—see the security overview and technical security documentation. However, we do not yet hold or publish SOC 2 or ISO certifications.
Ecosystem and community are early-stage, but the platform already includes runnable examples, integration paths, and implementation docs—start with the documentation.
The buyer category for a governed AI execution layer is still forming. PlanVault is currently in closed customer testing, and qualified teams can join with a free trial through early access.
We do not publish unverifiable named customers or unsupported production-scale claims; data handling expectations are described in the Privacy Policy.