Skip to main content

​
Security is the point, not a feature

You want the leverage of AI agents. What you shouldn’t have to accept is what a naive rollout requires: autonomous processes running with broad, long-lived credentials, next to your source code and customer data, with little visibility into what they touched or what it cost. Olyros Code removes that trade-off. You put agents to work inside your real repositories and systems without handing them the keys to everything: agents run contained, on infrastructure you control, with narrowly scoped credentials and a complete record of every action.

​
Where agents run

You manage agents from a central cloud console, but the work itself happens where your code and data already live.

On infrastructure you control

Agents execute in isolated containers on your own hardware or in your own cloud account. The heavy, sensitive work runs where you can see and govern it, not in a shared pool you cannot inspect.

Data locality by design

Your repositories and data stay next to the agent that needs them. The cloud service orchestrates and serves state; it is not where your code is checked out.

No inbound access required

Execution backends connect outbound only, like a self-hosted CI runner, so agents can run behind your firewall with zero inbound network configuration.

Resilient to outages

Work continues without connectivity and syncs on reconnect. Sandbox state stays accurate through network interruptions via redundant recovery paths.

​
Isolation and tenancy

Isolation is enforced at multiple layers rather than trusting a single check, and it is verified continuously.
Each orchestrator machine belongs to exactly one company. There is no shared execution host quietly running work for multiple customers.
An administrator must approve a machine before it can join the fleet and receive tasks. Nothing runs on unapproved infrastructure.
Every user is scoped to a single company at a time. That boundary is enforced both when a request arrives and again when data is queried, so a request for another tenant’s data is rejected outright rather than relying on a single gate.
An automated test suite walks the platform’s API surface to catch any endpoint missing tenant scoping, including attempts to guess another company’s identifiers. This makes it structurally hard for a new feature to leak data across tenants.
Access is revoked instantly by deactivating a membership or a machine, without deleting any data. Turning access back on is just as fast.

​
Credentials that stay small and short-lived

The biggest risk with agents is not the model; it is the credentials you hand it. Olyros Code is designed to keep that blast radius minimal.

Encrypted at rest

Customer secrets are encrypted with authenticated encryption before they ever reach the database, so any tampering with stored ciphertext is detected and rejected. Non-sensitive metadata is deliberately left in the clear; every actual secret is encrypted.

Short-lived and minimally scoped

Agents receive narrowly scoped, time-limited credentials for the task at hand, for example a one-hour GitHub token scoped to only the repositories that task needs.

Held in memory, never on disk

Secrets are injected into a container only at runtime and are never baked into an image or written to disk. They are masked in configuration previews.

Rotate once, centrally

Credentials live in one place. Rotate a key centrally and every agent that depends on it picks up the change, with no scattered copies to chase down.
This design extends to the people using agents, not just the agents themselves. When a skill is exposed as a simple tool in Slack, the person running it never sees the credentials underneath — capability is shared across the organization while the secrets stay in the platform, scoped and short-lived. The usual alternative is API keys in spreadsheets and scripts passed around in chat; this replaces it.

​
Enforcement at the container level

Guardrails that live only in application logic can be bypassed by the very thing they are meant to contain. Olyros Code pushes enforcement down to the runtime.
1

Read-only means read-only

When a task is granted read-only access to a repository, that restriction is enforced by the container runtime, not just checked in application code. The agent cannot write where it was never given write access.
2

Least privilege in the container

Agents run as a non-root user in a hardened runtime the platform owns and controls, rather than trusting a user-supplied image to behave.
3

Ephemeral by default

Workspaces are destroyed after each task. Each run starts clean, so nothing lingers between tasks.
4

Inspectable before it runs

Every container is described by a single declarative specification you can preview before anything executes, so you can see exactly what an agent will run.

​
Access and audit

Governance is only real if you can control who does what and reconstruct what happened.

Role-based access control

Admin, Editor, and Viewer roles define what each member of a workspace can do.

Enforced tenant SSO

Sign-in is delegated to standards-based single sign-on using OAuth 2.0 and OpenID Connect with signed tokens, cleanly separating who you are from what you are allowed to see.

Audit logging

Security-relevant actions are logged, and the full configuration of any run can be captured and reviewed afterward for post-mortem and audit.

Full cost and tool-call visibility

Every tool call an agent makes is visible, along with its token and dollar cost, so agent work is as accountable as any other line item.

​
An honest note on compliance

Olyros Code is built on defense-in-depth and auditability: contained execution, layered tenant isolation, encrypted secrets, least-privilege credentials, and standards-based authentication. We describe that architecture plainly so you can evaluate it on the merits. We do not claim certifications we have not earned. If your evaluation requires specific attestations or a deeper architecture review, book a demo and we will walk through the design with your security team directly.

​
Get started

Get Started

Create a workspace and put a governed agent to work in your own environment.

Book a Demo

Walk through the security architecture with us and get your questions answered.