Trust Center
Every action has a principal, a limit and a record.
This page explains how each Meshly product handles identity, access, limits, scanning, audit and isolation, and how you can reach us about security.
When you hand work to an agent or data to a platform, the responsibility stays exactly where it was. This page shows what the products enforce, so your review can start from facts.
Meshly Build
Agents act, and somebody stays accountable.
Sign in with the directory you already run. OIDC single sign-on works with your existing identity provider, so you handle joiners and leavers where you already handle them. Two-step sign-in is available on every workspace, and we recommend switching it on.
People and agents get the same access levels. Viewer, Maintainer and Admin are granted on the workspace, an organization or a project, and they are inherited below it. The Members and Access pages show who holds what, where it comes from and when it ends.
A person and an agent are different principals. People read, plan, discuss, capture and approve. Claiming, starting and completing work stays with the agent that does it, and every approval is recorded against whoever gave it: a person, or the reviewer agent when a project runs review in acting mode.
Connect your tools as yourself. Claude Desktop, claude.ai, Claude Code and Codex connect with your own sign-in and carry exactly your own access. You see what a connection may do before you approve it, and you can revoke it at any time.
Limits hold when nobody is watching. Each agent has monthly spend and token budgets, API keys expire and are rate-limited, and each assignee has one active task per project by default. Approval gates stand before agent work reaches main, and automated review starts in record-only mode.
You can enforce the security gate. Every review comes with a scan for committed secrets, vulnerable dependencies and risky code patterns, and it runs on your own Station, where the code already is. With the gate on, a change with a new critical or high finding cannot be approved until it is fixed or waived, and every waiver records its reason and who granted it.
Every action leaves a record you can query. Task moves, agent actions, approvals and policy gates land in a system audit log with a timeline view and an export. You can export security evidence per task or date range as JSON, SARIF and HTML.
One workspace cannot reach into another. Each workspace is kept separate on a shared instance, and dedicated single-tenant instances are available. On the workstation, every task runs in its own isolated workspace. Secret values are not stored in project documentation.
Your code stays where your code is. Station runs the agent on your own workstations, and only the record of the work reaches the server: tasks, verdicts, decisions, knowledge and audit. Meshly hosts the server by default, or you can run it on your own infrastructure, airgapped if you need to.
One level deeper
See how the gates work.
The Control sheet covers approval gates, automated review, budgets, access levels and per-agent model choice in detail.
Open ControlMeshly Data Stack
Your platform runs on your terms.
It runs on your infrastructure. Meshly Data Stack is deployed on Red Hat OpenShift or on other clusters and servers you control, inside the EU or inside an airgap. No telemetry calls home.
Every interface shares one identity. Keycloak single sign-on federates sign-in across the stack’s interfaces, and OPA policies decide what may be reached. Every call through the MCP gateway carries two credentials, one for the person who asks and one for the agent itself, and OPA policies use both to decide which data the answer may use and whether each field comes back in full, masked or empty.
Secrets live in an audited store. Services collect their credentials from OpenBao at startup, with TLS, auto-unseal and full audit logging, so no credential has to sit in a configuration file.
Row-level security applies where people read data. Superset dashboards apply row-level security on top of Trino federation, so a shared dashboard shows each reader only their own rows.
One level deeper
See what runs in the stack.
The Architecture sheet covers every layer and service, how the stack connects to your sources and how it is operated.
Open ArchitectureMeshly CLI
Your code stays inside your network.
You run the model. Meshly CLI works with any OpenAI-compatible endpoint you operate, such as your own vLLM server or the Meshly AI appliance, so your code never has to leave your network.
A sandbox and a permission gate protect your code. The agent reads and changes code inside a sandbox, and it asks before any action that needs permission.
Terminal work lands on the same board and in the same record. Meshly CLI connects to your own Meshly Build instance, so work done in the terminal is governed and recorded like any other agent’s.
One level deeper
Board, agent and model can all run on your hardware.
The Sovereign sheet shows the fully local set, with Meshly Build self-hosted, Meshly CLI and a model you run.
Open SovereignIndependent review
Outside reviewers check our engineering practice.
Meshly Build has been through two external security audits, and each was followed by hardening across the API surface, which shipped in Server 2.0.0.
Every feature runs in Meshly’s own production environment before it ships. We develop Meshly Build on Meshly Build, under the same gates and the same audit trail it gives you.
- Audits
- Two external security auditsMeshly Build, hardening shipped in Server 2.0.0
- Evidence
- Security evidence export per taskJSON, SARIF and HTML
Bills of materials
We document what is inside in a standard format.
Our software bills of materials are in CycloneDX JSON. We make each one for your deployment and your release, so we deliver them directly to customers and keep a single generic file off the public site.
| Product | SBOM | How you get it |
|---|---|---|
| Meshly Data Stack | SBOM: We produce it on request for the release you run, covering every container image in the stack. | How you get it: Ask us, and we deliver it directly to you. |
| Meshly Build | SBOM: The security scan in review produces one SBOM for every scanned commit. | How you get it: For your own code, it comes in Meshly Build with the security evidence export. For the product itself, we produce it on request. |
| Meshly CLI | SBOM: We produce it on request for the release you run. | How you get it: Ask us, and we deliver it directly to you. |
Source-code escrow is available on request for customers whose exit plan needs it.
Report a vulnerability
Found something? Tell us first.
Send the report to hello@meshly.fi with “Security report” in the subject, and the people who build the product will read it.
Include the product and version, what you found, the steps to reproduce it and the impact you expect. Give us a reasonable time to fix the issue before you make it public, and we will keep you informed until it is resolved.
Machine-readable contact: /.well-known/security.txt
Test in good faith. Use only accounts and data that are yours or that you have permission to use.
Do not harm others. You must not access other customers’ data, run denial-of-service attacks, attempt social engineering or carry out physical attacks.
Keep it between us until it is fixed. Share the details only with us until the issue is resolved and we have agreed on disclosure.
Questionnaires
Security review
Send us your questionnaire.
If your review has a form, send it to us. The people who built the product will answer it, and you will get a direct answer on any specific control that matters to you.