Procurator is Mission Control for AI
Accelerate Autonomy. Enforce Governance. Optimize Spend.
Orchestrate multi-agent workflows with deterministic constraints and Human-In-The-Loop approvals. Monitor all inputs and outputs in real-time.
Enforce strict Row-Level Access Controls (RLAC) across all autonomous agents. Expose zero private keys to external clients.
Set precise token budgets. Automatically halt runaway processes before you accrue unexpected cloud charges.
Seamlessly mount tools to your Spring AI Agents via MCP Servers natively supported within the Procurator monolith.
Up and Running in Three Steps
No managed SaaS. No vendor lock-in. Your environment, your rules.
Install in Your Environment
Stand up Procurator on-premise, in a private cloud, or air-gapped. One Spring Boot JAR, one PostgreSQL database. Your data never leaves your perimeter.
Wire Up Agents & Tools
Register your LLMs, mount MCP tool servers, configure agent teams and DAG workflows. Agents are fully functional from day one — no framework glue code required.
Monitor, Control & Sleep Well
Set HITL approval gates, token budget caps, and RLAC policies. Watch every run in real-time. Every action is logged to an immutable audit vault — nothing goes untracked.
Built for Your Entire Organization
One platform. Three teams that ship AI faster and sleep better.
Stop gluing frameworks together. Procurator gives you a native Spring AI runtime, hot-reload model switching, and a live MCP tool registry — so you ship production agents, not plumbing.
- Agents fully functional Day 1 with native Java tools
- Python sandbox & webhook extensions without recompile
- Visual DAG builder for deterministic multi-step workflows
- Evaluation harness to benchmark against ground-truth datasets
Every tool call, every retrieval, every model response — intercepted, logged, and governed. Procurator enforces your policy before any action reaches production.
- 3-tier HITL approval gates that pause and resume agent threads
- RLAC row-level access control across all vector retrievals
- PII anonymization & prompt injection defense built-in
- Immutable audit vault — full prompt-path payload with token trail
No surprise cloud bills. No black-box inference. Set hard token budgets per agent, get real-time burn dashboards, and kill runaway processes before they cost you.
- Hard budget caps — automatic halt before spend limits breach
- Per-tenant cost attribution with FinOps gateway dashboards
- Emergency kill-switch for runaway inference chains
- Deploy air-gapped or on-premise — your data never leaves your perimeter
Platform at a Glance
Six integrated control suites covering every dimension of production agentic AI.
Design, deploy, and operate autonomous agent workforces — from single-agent loops to coordinated multi-agent teams and deterministic DAG workflows.
- Agents — Configurable LLM-backed agents with dynamic tool stacks
- Teams — Coordinate, Route, Broadcast & Debate topologies
- Workflows — Visual DAG execution with deterministic step ordering
- Evaluations — Benchmark agent quality against ground-truth datasets
Manage the full compute layer: hot-load LLMs from any provider, expose and consume MCP tools, and register dynamic agent extensions without recompilation.
- Models — Multi-provider LLM registry with hot-reload
- MCP Servers — Native client + expose as MCP Server for IDEs
- Tool Registry — Dynamic JSON tool mounting, no recompile needed
- Extensions — Java SPI, Python sandbox & webhook hooks
Ground agents in your private data. Drag-and-drop RAG ingestion with pgvector HNSW indexing and governed long-term semantic memory — all within your perimeter.
- Knowledge — RAG upload center with live vector chunk inspector
- Memory — Long-term user memory with compaction & deletion controls
- RLAC — Row-level access enforcement across all vector retrievals
- Semantic Search Bench — Graphical retrieval quality test bench
Full operational visibility into every running agent. Track sessions, manage background runs, approve HITL checkpoints, schedule recurring routines, and federate across nodes.
- Sessions — Time-indexed conversation history explorer
- Runs — Background worker swarm tracking dashboard
- Approvals — 3-tier HITL checkpoint intercept & resume
- Schedules — Cron-based autonomous agent routines
- A2A Mesh — Agent-to-agent federation across nodes
End-to-end observability from TTFT to full workflow latency. Real-time FinOps dashboards with per-tenant cost attribution, budget enforcement, and an emergency halt switch.
- Observability — OpenTelemetry traces hooking into Prometheus
- FinOps & Gateway — Token burn rates, cost dashboards, budget caps
- TTFT Metrics — First-token to full-response latency breakdowns
- Real-time Halt — Emergency kill-switch for runaway inference chains
Proactive alerting on budget overruns and policy violations. Enterprise user management, role governance, and a full immutable audit trail vault.
- Alert Rules — Threshold-based triggers with integration webhooks
- Budget Alerts — Automatic notifications before spend limits breach
- Security & Audit — PII anonymization, prompt injection defense
- Audit Logs — Full prompt-path payload vault with token footprint tracking
Ready to take control?
Deploy Procurator in your own environment. Air-gapped, on-premise, or private cloud — your data never leaves your perimeter.
Built With