AI Agents
AI agents as accountable members of your organization
Envoy makes AI agents first-class members of your organization, holding the same roles, permissions and team memberships as people but never able to sign in interactively. IT and security teams get one place to issue, scope, budget, monitor and shut down every agent, inside Neuraforge or on an external framework. Every tool call is audited, sensitive actions can require human approval, policy decides who may read an agent's conversation, and one switch stops the fleet.
Most agent platforms bolt identity onto each tool, so agents run on shared keys and nobody can say who did what. Envoy gives every agent its own identity on the same permission engine as your people, routes its model calls through governed dispatch, audits every tool call, and lets one switch stop the whole fleet.
- Synthetic agent identities with full roles, teams, and permissions
- Every agent model call routes through governed AI dispatch
- MCP gateway exposes platform APIs, and your functions, scripts and workflows as tools, to external AI frameworks
- Action audit log of every tool call, response and decision, with approvers opening the run's conversation by policy
- Action control — caps, time-of-day limits, approval gates, and an instant kill switch
What you get
What Envoy gives your team.
Agents as organization users
Agents hold roles, permissions and team memberships evaluated by the same engine that governs people, with a fixed set of powers no agent can ever hold.
Personal and organization agents
Give each person an AI twin whose authority never exceeds theirs, or run autonomous organization agents that always have an accountable human owner.
Works with any agent framework
An OpenAI- and Anthropic-compatible endpoint means frameworks connect by swapping a base URL, with each agent governed, metered and audited by identity.
Cost caps and a kill switch
Set daily and monthly cost caps and rate limits per agent, and stop every agent in one action. Disabling takes effect on the agent's next request.
Tool profiles without exposed secrets
Grant tools through named, versioned profiles bound to agents, teams or roles. A waiver names its tools and a remembered approval is tied to its server, so new tools ask first.
Approval and action control
Require human approval for sensitive tools; an agent's current owner decides and the starter hears every outcome. Action policies govern what an agent may do, with an append-only decision record.
How it works
- 1
Create the agent
Add a personal AI twin whose authority never exceeds its person's, or an autonomous organization agent with an accountable human owner. Give it roles and team memberships like any employee.
- 2
Issue its delegated credential
Each agent gets one credential, shown once and revocable on demand, with an optional expiry. Point an agent framework at the compatible API by swapping its base URL; the same key reaches tools over MCP.
- 3
Grant tools, limits and approvals
Bind tool profiles to the agent, set daily and monthly cost caps and rate limits, write action policies, and require human approval for sensitive tools.
- 4
Monitor, quarantine or stop
Review each agent's usage, cost and approvals in the registry. Quarantine routes future actions to a person for review; disabling takes effect on the next request, and one switch stops every agent.
Ask AI about Envoy
Every Envoy page has an Ask AI button. It opens a chat that already has the page, its rows, and your selection. Try asking:
- Which agents on this page are disabled, and which expire soonest?
- Which agents here have the highest cost this month?
- What does this action policy refuse, and is it in audit-only mode?
How teams use it
- Let a personal AI assistant act for an employee without exceeding that person's access.
- Run an autonomous support agent with a monthly spend cap and a named human owner.
- Quarantine a misbehaving agent for review without halting work already in progress.
Frequently asked questions
See Envoy in action
Join the early-access list, or talk to our team about your environment.
Talk to us