Crescendo Lab

Agent Lab Platform · Enterprise AI agent platform

An enterprise AI agent platform IT can sign off on.

Agents use the asking person's own permissions, stop at a spending cap, and leave a record of every call and approval.

Still comparing? See how to choose.

The basics

What is an enterprise AI agent platform?

An enterprise AI agent platform is the system a company uses to build, launch and manage AI agents: who may use which agent, what data it may touch, what it may spend and what it did are all set and checked in one place.

Agent Lab
An enterprise AI agent platform from Crescendo Lab.
What it does
People hand work over in the web console or Slack; agents look things up and use company systems to finish it.
Permissions
An agent can only use the permissions the asking person already has.
Spend
Workspaces and teams have spending caps; calls past them are refused with the reason.
Records
Every model call, approval and credential use is logged and can be exported.

Types

What types of AI agent platforms are there?

Roughly four. They differ in who builds the permissions, budgets and records.

TypeCommon examplesFitsWhat IT still has to add
Developer frameworksLangGraph, CrewAI, OpenAI Agents SDKTeams with engineers who want full controlMost of the interface, permissions, approvals, budgets and records.
Low-code and open-source platformsDify, n8n, BotpressA quick proof of conceptPermissions, audit and spending caps vary by plan; self-hosting means running it yourself.
Big-vendor suitesMicrosoft Copilot Studio, Gemini Enterprise, ChatGPT Enterprise, Claude EnterpriseCompanies whose data and work already live with that vendorYou follow that vendor's permission settings, pricing and data regions.
Agent LabCrescendo Agent LabCompanies rolling agents out to every team, with IT in charge of access and costPermissions, approvals, budgets and records are built in; agents written in a framework get budgets and records through the AI Gateway.

Examples are common products; features vary by plan.

How to choose

How to choose an enterprise AI agent platform: six questions for IT

  1. Could an agent see data it should not?

    Check: Whether the agent acts with the asker's own permissions or with one powerful shared account.

    Agent Lab: The asker's own permissions; knowledge search runs as them too.

  2. Can a person approve before an agent acts?

    Check: Whether sending email or changing records can pause until someone approves.

    Agent Lab: It pauses by default, then runs after approval in the console or on a Slack card.

  3. Can we cap the spend?

    Check: Whether the cap is an alert after the fact or a block before the call.

    Agent Lab: Cost is reserved before the call; past the cap the call is refused with the reason.

  4. If something goes wrong, can we see who did it?

    Check: Whether records name the person, agent, model and cost, and can be exported.

    Agent Lab: Every call and approval is recorded; owners can export all of it.

  5. Where does our data live?

    Check: Ask about platform data and model inference separately; they are often in different places.

    Agent Lab: Cloud, hybrid or on-premises; inference depends on the model you pick.

  6. Can it govern agents built with other tools?

    Check: Whether it only covers its own agents, or agents written in a framework can join.

    Agent Lab: Agents that call models through the AI Gateway get spending caps, records and a model allowlist.

Architecture

What is Agent Lab made of?

People hand work over in Slack or the web console; four modules handle where agents are used, which model they call, which company systems they reach and where they run; one layer of identity, policy, approval, usage and audit sits under every agent.

Agent Lab Platform

A secure, governed, and auditable enterprise agent platform

The four products share one identity, policy, approval, usage, and audit layer.

effective = user ∩ assistant ∩ toolUser's own permissionwhat they already see in ERPAssistant's capabilitywhich tools it was grantedTool's own permissionwhat the credential allowsEffective access

Delegated identity:
assistants act as the person asking

A tool call runs only if the user, the assistant, and the tool all allow it. Your existing permissions are the rules.

  • Knowledge retrieval runs as the asker, so hidden documents are never cited
  • Connectors use a shared organization connection or the member's own
  • Permission is rechecked before execution, so removed members stop acting
Model proposesa tool callThree checkstool · rules · screeningWaits for a personawaitingApprovalNeeds your approvalWants to call erp.inventory.lookupRuns with Sam Lin's permissionsApproveDenyRuns and is recordedwho asked · who approved · as whomApprove in the console or on a Slack cardRead-only tools an owner marksas automatic can skip the wait

Approval where work happens:
tool calls pause for a person

By default, tool calls wait for approval in the console or on a Slack card, then run once after a permission recheck.

  • Business rules can require approval for a tool, or forbid an action
  • When screening escalates a turn, all its tool calls need approval
  • Read-only tools an owner marks automatic can skip the wait; one switch turns it off
Workspace monthly budget62%80%Under 80%the model that was chosenauto picks by difficulty80% to the capowners are alertedauto skips the strongestPast the ceilingrefused with a reasonnever borrows next monthLedger · one row per model call10:42 procurement · sam.lin · claude-opus · 3,120 tok · chat10:43 procurement · sam.lin · claude-opus · 1,840 tok · tool11:00 weekly-report · auto · claude-haiku · 22,400 tok · schedule11:05 kb-index · system · claude-haiku · 8,900 tok · index

Budgets that hold:
ceilings per workspace and per team

Reserve the budget, call the model, write the ledger. At a spend ceiling, a model the system picked steps down to a cheaper one and says so; a model the member chose is never swapped; past the cap, calls are refused with the reason.

  • Monthly, rolling-window, and per-team ceilings
  • Each call logs who, which assistant, which model, tokens, and why
  • With auto, the model follows task difficulty: cheap for simple work, strong only for multi-step tasks

Platform Security

What IT checks before go-live is already built in

Data residency, sign-in, and incident tracing are set up before go-live.

Data and isolation

  • One tenant boundary per workspace
  • Optional dedicated Cloud SQL database per workspace
  • Credentials sealed with AES-256-GCM, short-lived grants per use
  • On-premises systems connect through a relay, so data need not move to the cloud (compare cloud, hybrid and on-prem)

Identity and audit

  • SAML and OIDC SSO, directory groups mapped to roles
  • Delegated identity, approvals, and token budgets in every workspace
  • Approvals, membership changes, and credential use logged and exportable
  • Slack, Teams, and LINE share one identity, budget, and audit trail

Cloud and on-prem

Runs where you need it: cloud, hybrid or on-premises

Same assistants and governance; where your data may go decides where it runs.

Cloud

Hosted by us, one tenant per workspace, no machines to prepare.

For teams starting this week whose data can live in the cloud

Hybrid: platform in the cloud, data on your network

Systems behind your firewall connect through a relay you run: it dials out with no inbound rule, credentials stay on your network, and queries are filtered to the asker's rows.

For companies with core systems in their own data center

On-premises: the whole platform in your environment

Runs in your data center or private cloud; data and audit records stay inside.

For finance, healthcare and public-sector data that must stay inside

FAQ

Enterprise AI agent platform questions

How is an enterprise AI agent platform different from tools like ChatGPT?
A personal AI tool helps one person; an enterprise AI agent platform lets IT decide who in the company may use which agent, which systems it may touch and what it may spend, with every step on record.
Can Agent Lab govern agents built on other platforms?
Partly: an agent written in any framework gets spending caps, a record of every call and a model allowlist once it calls models through the AI Gateway; business rules, the content screening before tool calls and evaluations currently apply only to agents built in Agent Lab. See the AI Gateway
Can we use Agent Lab without the cloud?
The cloud edition gives each workspace its own tenant; where inference runs depends on the model you pick. Connecting on-premises systems or installing the whole platform on-premises is scoped per project.
Does it work with our existing SSO?
Yes. SAML and OIDC single sign-on, with directory groups mapped to workspace roles.
Can we try it first?
Yes. Sign up to open a workspace yourself, or book a demo to run approvals and budgets on one of your own scenarios.

Sign up free, or book a demo

  • 45 minutes on an approval and a budget in your scenario
  • On-prem or hybrid? We'll look at your environment
  • Straight answers on data, permissions, and cost
  • If it fits, a workspace the same week
Book a demo

Optional, helps us schedule faster