Permit.io provides fine-grained authorisation as a service for applications, APIs and AI agents. It is built for engineering teams that want to stop hard-coding permission logic into their software: it sits behind the existing identity provider and decides what each human user, service or agent is allowed to do.
The problem it solves grows with AI: agents now act on behalf of users, and each of their actions needs the same permission check a person’s request would get.
At A Glance
| Vendor | Permit.io |
|---|---|
| Category | Authorisation |
| Identities covered | Human users, services and AI agents |
| Deployment | Cloud, hybrid and on-premises, with a local policy decision point |
| Agentic AI | MCP Gateway, delegated agent access, AI Access Control |
| Customers | Maricopa County Recorder Office, Honeycomb Insurance, Centauri AI |
| Certifications | SOC 2 Type II, HIPAA BAA available, GDPR, CCPA |
| Funding | Series A (2024) |
| Market classification | Authorisation |
What Permit.io Does
Permission checks scattered through application code are hard to change, audit or extend. Permit.io moves that logic into a service. Decisions run at low latency in a policy decision point inside the customer’s own network, while policies are managed centrally through a no-code editor, an API, a Terraform provider, a CLI and SDKs.
The same engine now covers AI agents. Agent access is treated as delegation from a person: an agent receives a subset of a user’s permissions, “never a standing god key”. Agents are then checked against policy in the same way as a request from a user.
Products
App And API Permissions
App and API Permissions is the core product for developers adding roles and fine-grained rules to their own software.
- RBAC, ABAC and ReBAC access models.
- Policies in OPA Rego or AWS Cedar, kept in Git, with a no-code editor that writes the policy code.
- Every decision logged with an explanation, and forwardable to a SIEM.
- Support for API gateways such as Kong.
Nexus PDP
Nexus PDP is a newer policy decision point, in early access, for applications where authorisation latency matters. It stores the authorisation graph on disk in SurrealDB, receives changes over a durable NATS stream and evaluates policy with OPA without any network hop in the decision path. A reported 5.6 ms P99 for an RBAC check was measured server-side on a single 2 CPU pod.
MCP Gateway
Permit MCP Gateway sits between AI agents and MCP servers, for organisations that want agents to use tools without handing them raw credentials.
- OAuth 2.1 for MCP clients, with sign-in through SAML or OIDC SSO, Google, GitHub, Microsoft, email or passkeys.
- A policy check on every tool call, consent screens and tools classified by risk.
- Agent fingerprinting that flags drift in behaviour, and intent checks that deny actions straying from the delegated job.
- Every allowed or denied call logged with the human, agent, tool, server, time and reason.
AI Access Control
AI Access Control protects AI features inside applications with four perimeters: prompt filtering, protection of data retrieved for RAG, secure access to external systems and enforcement on responses. It analyses the whole request chain, so each action can be traced to its human or AI source, and critical actions such as purchases can require human approval.
Elements
Elements are embeddable permission components that can be placed inside an application, so end users can manage access themselves without the team building those screens.
Open Source
| OPAL | An administration layer for Open Policy Agent that detects changes in policies and policy data in real time. |
|---|---|
| OPToggles | Pairs OPA and OPAL with feature flag tools so that frontends reflect backend permissions. |
How It Works
Permit.io separates managing policy from making decisions. Deployment can be cloud, hybrid or on-premises, and sensitive data can stay in the customer’s VPC.
| Component | What it does |
|---|---|
| Control plane | Policies, roles and relationships are managed in the Permit.io service through the editor, API, CLI or Terraform. |
| Local PDP | OPA or Cedar runs in the customer’s VPC. OPAL keeps each PDP in sync with policies and data from sources such as APIs, databases, Git and S3. |
| MCP Gateway | Agents hold no upstream credentials: the gateway stores them and issues its own tokens. When a person is removed or withdraws consent, the agent loses access on its next call. |
Identities And Systems Covered
| Human users | Application end users authenticated through an existing identity provider. |
|---|---|
| Services | APIs and microservices that call the check API. |
| AI agents | Agents calling MCP servers, each bound to the human who delegated to it. |
| Data | RAG queries and vector database entries filtered by attribute-based policy. |
Integrations
| Area | Integrations |
|---|---|
| Policy engines | Open Policy Agent (Rego), AWS Cedar |
| AI frameworks | LangChain, PydanticAI, LangFlow |
| MCP clients | Cursor, Claude Desktop, VS Code, Claude Code |
| Infrastructure tooling | Terraform |
Company Background
| Founded | 2021. |
|---|---|
| Headquarters | Tel Aviv, Israel. |
| Leadership | Or Weis, Co-founder and CEO; Asaf Cohen, Co-founder. |
| Funding | $8 million Series A in 2024, led by Scale Venture Partners with NFX, Verissimo Ventures, Roosh Ventures, Firestreak and 92712. |
Note: This overview is compiled from Permit.io’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.