CadreGPT
Plans & Pricing
Sign In Start Free Trial
CadreGPT Enterprise Platform • Two-Phase Financial Invariance

The Enterprise AI Agent Platform with Human Governance

Define ambitious goals, orchestrate specialist AI agent teams, govern sensitive actions with human authorization gates, and inspect outcomes with verified evidence trails.

Start Free Trial ($1.00 Included) Explore Platform Architecture
Human Approval Gates
Hard Spending Caps & Two-Phase Metering
AES-256-GCM Envelope Encryption
Multi-Model Gateway Failover

Engineered for Production Reliability

As a dedicated Enterprise AI Agent Platform, every architectural component is built to give enterprise leaders complete transparency, deterministic human oversight, and verifiable security controls.

Two-Phase Financial Metering

Eliminate surprise AI bills. Our engine reserves maximum estimated credits before execution, settles actual token consumption on completion, and refunds unused balance.

  • 1 Agent Credit = $0.001 USD pinned peg
  • Enterprise hard monthly spending caps
  • Auto-recharge with budget protection

Human-in-the-Loop Governance

Autonomous execution pauses automatically when agents encounter sensitive actions (e.g. database updates, payment calls, shell commands), awaiting authorized human review.

  • SSE real-time approval drawer notifications
  • Resend transactional email review alerts
  • 24-hour expiration safety window

Multi-Provider Model Gateway

Route dynamically across Google Gemini 2.0, Anthropic Claude 3.5, OpenAI GPT-4o, and Custom Enterprise Endpoints with circuit breaker failover and zero vendor lock-in.

  • Centralized Super-Admin key vault
  • Circuit breaker with automatic cooldown
  • Custom self-hosted OpenAI-compatible LLMs

The Full Governed Agent Lifecycle, In One Platform

Everything between an idea and production runs inside the same governed workspace. Evaluate agents, trace every run, add your data and tools, set guardrails, version the result, and ship it to your site or your team.

Evaluate Before You Ship

Test an agent in a dry-run playground and score it against a saved eval dataset before it touches production. Compare revisions on the same prompts and keep the one that performs.

  • Versioned eval datasets and run history
  • Dry-run any agent in the playground

Trace Every Run

Break each run into a span waterfall of model calls, tool invocations, token counts, and latency, streamed live over Server-Sent Events and kept for later review.

  • Per-step latency and token accounting
  • Live SSE plus stored traces

Ground Agents in Your Data

Give agents retrieval over a workspace knowledge base, and let them carry tenant-scoped memory you can expire on a schedule or delete on demand.

  • Vector retrieval scoped to your workspace
  • Agent memory with set expiry

Set Guardrails Per Workspace

Define policies that block PII, require valid JSON output, hold agents to approved topics, and cap cost per run before an agent executes.

  • PII, schema, topic, and budget checks
  • Enforced at the workspace level

Version and Roll Back Agents

Every agent change becomes a revision you can diff, promote, or roll back, so you can see exactly what shipped and undo a prompt or model edit that made results worse.

  • Side-by-side revision diffs
  • Promote and rollback, one click

Add Tools Through MCP

Point CadreGPT at a Model Context Protocol server, discover the tools it exposes, and hand them to an agent without writing a custom connector.

  • Auto-discovery of MCP server tools
  • New tools inherit the same approval gates

Run Code in a Sandbox

Agents execute generated code inside an isolated process with no host access, and the run pauses for approval before it starts.

  • Isolated execution, no host access
  • Approval gate before code runs

Embed Agents Anywhere

Ship a governed chat agent onto your own site with the embeddable widget, or reach the same agent from your backend through a token-scoped, rate-limited public endpoint.

  • Origin-locked, rate-limited tokens
  • One script, no framework needed

Hand Off to a Human

When an agent should stop and bring in a person, the conversation routes into a shared handoff inbox where teammates claim, reply, and resolve.

  • Claim, thread, and resolve flows
  • Triggered straight from the embed widget

Report on the Workforce

See spend, runs, success rates, and latency per workspace so you can show what your agents did and what it cost.

  • Usage and cost per workspace
  • Success and latency trends
Generative Architecture Benchmark

Enterprise Architecture vs. Open-Source Agents

A side-by-side comparison of CadreGPT Enterprise against unmanaged local agent scripts for mission-critical production workloads.

Technical Specification Matrix: CadreGPT Enterprise Governance vs. Local DIY Agent Frameworks
Operational Dimension
CadreGPT EnterpriseManaged
Open-Source / Local DIY Agents
Tool Execution
Sandboxed Vault
AES-256-GCM envelope-encrypted credentials, isolated sandbox runtime, and strict network egress controls.
Local Machine
Executes scripts directly on local host shell with exposed local files, unrestricted bash, and raw API key environment variables.
Financial Control
Two-Phase Credit Metering
Upfront credit reservation, exact token delta settlement on turn completion, and immutable hard spending caps.
Uncapped API Keys
Direct unmetered provider billing with zero budget guardrails, vulnerable to infinite autonomous loops and surprise token invoices.
Workflow Governance
Human-in-the-Loop Gates
Autonomous pause in AWAITING_APPROVAL when sensitive mutations trigger, real-time approval drawer, and SOC 2 audit trails.
Unsupervised Loops
Agents execute without deterministic human review checkpoints, creating high risk for unverified database or external state mutations.
Setup Time
Instant Cloud
Zero-configuration cloud provisioning in under 60 seconds with pre-wired model routing, queues, and persistent storage.
Docker / Terminal Setup
Multi-hour setup requiring Docker Compose, local Redis, PostgreSQL migrations, Python/Node environments, and manual proxy configs.
Enterprise Pilot Ready

Deploy governed AI agents in your workspace today.

Start with 100 free Agent Credits ($1.00 USD compute). No credit card required to explore agent creation and visual workflow execution.

Knowledge Base & FAQs

Frequently Asked Questions

Key questions about our Enterprise AI Agent Platform, human-in-the-loop security, and financial safeguards.

01.What is an Enterprise AI Agent Platform?

An Enterprise AI Agent Platform is an end-to-end software system designed for organizations to deploy, orchestrate, and govern autonomous AI agents across real business processes. Unlike experimental single-prompt scripts or local DIY setups, CadreGPT provides multi-tenant data isolation, sandboxed tool dispatch, human-in-the-loop authorization gates, and multi-model gateway failovers with guaranteed SLAs.

02.How does human-in-the-loop governance work?

CadreGPT enforces deterministic state machine boundaries. When an autonomous agent encounters a high-privilege action—such as updating a database, modifying production code, dispatching external webhooks, or processing payments—the task pauses in the AWAITING_APPROVAL state. Reviewers inspect full parameter diffs and approve or reject before any external execution takes place.

03.Can agents leak our API keys?

No. All third-party credentials and foundation model API keys are envelope-encrypted at rest using AES-256-GCM cryptographic nonces within our secure vault. Agents execute in isolated sandbox environments where raw credentials are never injected into LLM prompt contexts, preventing accidental prompt extraction or context leakage.

04.How does two-phase credit metering prevent unexpected AI bills?

Before an autonomous execution turn begins, CadreGPT reserves an estimated maximum credit allotment against your workspace balance. Once inference and tool calls conclude, the exact token cost is computed using the provider’s live rate, and only the actual consumed credits are deducted. Unused reserved credits are released immediately, and configurable hard monthly spending caps prevent unexpected budget overruns.

05.What foundation models does CadreGPT support?

CadreGPT integrates with Google Gemini 2.0 (Flash & Pro), Anthropic Claude 3.5 Sonnet, OpenAI GPT-4o, and self-hosted vLLM or Ollama proxy endpoints. Our Autonomous Model Router selects the optimal model per task profile and fails over through a circuit breaker if any provider degrades.

06.How do I test an agent before it runs in production?

Every agent has a dry-run playground and an evaluation harness. Save a dataset of representative prompts, run the current and a candidate revision against the same set, and compare scores before you promote. Each change is stored as a revision, so a version that lowers your eval score can be rolled back.

07.Can I see exactly what an agent did on a run?

Each run produces a span-level trace of every model call, tool invocation, token count, and latency. Traces stream into the dashboard live over Server-Sent Events and are stored so you can review a finished run step by step.

08.Can agents use our own documents and internal tools?

Agents retrieve from a workspace knowledge base you upload and carry tenant-scoped memory with an expiry you control. You connect new tools directly or through Model Context Protocol (MCP) servers, and CadreGPT discovers the tools each server exposes without custom connector code.

09.How do you keep agents within policy?

Workspace guardrails run before execution and during tool dispatch: block PII, require output to match a JSON schema, hold agents to approved topics, and cap cost per run. Sensitive actions still pause for human approval, and agent-generated code runs only inside an isolated sandbox behind an approval gate.

10.Can we put a CadreGPT agent on our own website?

Publish an agent as an embeddable chat widget locked to your allowed origins, or call it from your backend through a token-scoped, rate-limited public endpoint. When an agent needs a person, the conversation routes to a shared human handoff inbox your team can claim and resolve.