Oso builds authorisation for software teams: a central service where engineers define who can do what in their applications, instead of hand-coding permission checks throughout the codebase. It has since extended that work to AI agents, with tools that inventory agents, watch what they do and enforce policy on their tool calls. “Agents are here. Oso makes them safe.”
It serves two audiences: engineering teams who want permissions handled by a dedicated service, and security teams who need to know which AI agents are running, whose authority they act with and what they touch.
At A Glance
| Vendor | Oso |
|---|---|
| Category | Authorisation |
| Identities covered | Application users, roles and permissions; AI agents |
| Deployment | Oso Cloud (managed service); Oso Self-Hosted on AWS (private beta) |
| Agentic AI | Oso for Agents; Oso for Coding Agents; Oso MCP Server |
| Customers | Brex, Duolingo, Webflow, Wayfair, PagerDuty |
| Certifications | SOC 2 |
| Market classification | Authorisation |
What Oso Does
Authorisation logic is hard to build and even harder to change once it is spread across application code. Oso is meant for engineering teams that are done rolling their own permissions: they write the permission model once in a policy language and query it from their services. That makes it easier to simplify role management, decouple access control from the application and keep permissions consistent while moving from a monolith to microservices.
The same thinking now extends to AI agents, which act with real user permissions and call tools on their behalf, so they need the same kind of policy and oversight as the people they work for.
Products
Oso Cloud
Oso Cloud is the managed authorisation service. Developers express any authorisation model, from role based to attribute based, in Polar, a domain specific language for permissions logic, and the application asks Oso Cloud for decisions at runtime. The service is built in Rust, runs across more than 30 availability zones and scales horizontally, with a reported p90 latency under 10ms. Local Authorization, Fallback and Oso Sync are generally available, and Oso Migrate helps teams move existing authorisation logic onto Oso. Developers building with Oso can also use the Oso MCP Server as an authorisation co-pilot.
Oso Self-Hosted
For organisations that need authorisation to run inside their own cloud account, Oso Self-Hosted is available in private beta. It runs on AWS, scales horizontally on ECS and is managed with an oso-manager command-line tool.
Oso For Agents
Security teams often do not know which AI agents are running, what they connect to or whose authority they use. Oso for Agents answers that with discovery, monitoring, detection, control and reporting. It inventories agents, the connectors and tools they call and the users and groups behind them, alerts on unsanctioned agents or sensitive data in Slack, and applies policies that block a tool call, require approval or simply alert the team. For monitored agents every prompt, completion and tool call is captured, and default detections cover rule violations, high-velocity behaviour, PII leakage and exposed API keys, with room for custom rules such as blocking unknown MCP servers.
Oso For Coding Agents
Oso for Coding Agents, in early alpha, gives engineering and security leaders a view of every coding agent in the organisation, the tools it calls and the data it touches. Each tool call receives a risk score based on frequency and severity, so teams can see which agent behaviour deserves attention first.
How It Works
| Component | What it does |
|---|---|
| Oso Cloud | Deployment and scaling are handled by Oso. Applications send authorisation requests to the service, which evaluates Polar policies, and the production architecture isolates customers and their data from each other. |
| Oso Self-Hosted | Two services run in the customer’s AWS account. The Oso Service handles traffic from application clients, with MSK (Kafka), S3 and SQS as its main dependencies, and the Oso Snapshotter generates snapshots. Nothing else needs to be exposed outside the network. |
| Oso for Agents | Agent activity is discovered across endpoints, browsers and network traffic, and policies are enforced centrally at the network level. Oso for Coding Agents connects through AI gateways that sit between developers and LLM providers. |
Identities And Systems Covered
| Application users | Users, roles and permissions modelled in Polar, including role based and attribute based models. |
|---|---|
| AI agents | Agents, their connectors and tools, and the users and groups behind their activity. |
| Coding agents | Claude, OpenAI, Gemini, Codex, Cursor, GitHub Copilot and Antigravity. |
| Identity sources | Okta, connected through Identity Sync. |
Integrations
| Area | Integrations |
|---|---|
| Security operations | Splunk, CrowdStrike, SentinelOne, Jamf, Elastic |
| AI gateway | Tailscale Aperture |
| Alerting and storage | Slack, Amazon S3 |
Company Background
| Headquarters | New York, United States. |
|---|---|
| Leadership | Graham Neray (CEO and Founder), Nick Kucharski (CTO). |
| Investors | Include Bill Coughran of Sequoia Capital and Jake Storm of Felicis Ventures. |
| Acquisition | In October 2026 Oso announced that Oso Cloud had been acquired by Tiny and would continue to operate as usual, while most of the team moved to Temporal to build authorisation for AI agents. |
| Recognition | Finalist in InfoWorld’s 2025 Technology of the Year Awards for Application Security. |
Note: This overview is compiled from Oso’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.