Security

Agents act. Somebody has to be accountable for it.

Handing work to an agent does not move the responsibility anywhere. Meshly Build is built so that every action has a principal behind it, a limit around it, and a record after it. This page covers how that works, and what we do not claim.
SECURITY POSTURE
Who is acting?
SSO · two-step sign-in
What may they do?
Access levels · scoped connections
What did they do?
Audit log · queryable timeline
How far can it go?
Budgets · approval gates
01 · Identity and access

Sign in with the directory you already run.

Meshly Build signs people in through OIDC single sign-on, so joiners and leavers are handled where you already handle them. Access levels decide what each person can reach: Viewer, Maintainer or Admin, granted on the workspace, an organization or a project, and inherited by everything below. Agents get access exactly the same way.

  • OIDC single sign-on with your existing identity provider.
  • Viewer, Maintainer and Admin access levels, inherited from workspace to organization to project. People and agents alike.
  • Members and Access pages at every level show who holds what, where it comes from, and when it ends.
  • Exclude someone from a single project without touching their wider access, and let project Maintainers run their own projects.
ACCESS LEVELS · WORKSPACE · ORGANIZATION · PROJECT
Granted where it is needed. Inherited below.
Admin
Settings, people, agents and access
Maintainer
Runs its projects: members, agents, budgets
Viewer
Reads what the grant covers
People and agents get access the same way.
02 · Two-step sign-in

A second step for every person who signs in.

Two-step sign-in adds another layer of protection to every workspace, so a password alone is never enough. It is optional, and we highly recommend switching it on.

  • Available on every workspace, and highly recommended.
  • Agents and integrations keep working as they do.
TWO-STEP SIGN-IN
A password alone is not enough.
Every workspace
available today
Our advice
switch it on
Agents + integrations
unaffected
03 · Principals

A person and an agent are not the same principal.

Someone planning in a chat window and an agent executing on a workstation get different tool profiles, on purpose. People read, plan, discuss, capture and approve. Claiming, starting and completing work stays with the agent that is actually doing it. That separation is what makes the audit trail mean something afterwards.

  • People get a conversational profile. The execution verbs are not in it.
  • Agents get access through the same access levels as people, so an agent sees and changes only what its level allows.
  • Approving your own agent’s work is a human action, recorded against a human.
SESSION PROFILES
Person
·read·plan·discuss·capture·approve
Agent
·claim·start·edit·complete
A chat window cannot claim a task. That is deliberate.
04 · Authorization

Connect your tools. As yourself.

Plan from Claude Desktop or claude.ai, or work from Claude Code or Codex, and you sign in as yourself, carrying your own access into the session with no shared key to hand around. You see what a connection may do before you approve it, and read-only access is read-only.

  • Everyone signs in as themselves, with their own access.
  • Connections are scoped, and you can revoke them at any time.
  • A connection carries your own access, and nothing beyond it.
CONNECTED TOOLS
Nothing is granted before it is named.
TOOLS
Claude Desktop · claude.ai · Claude Code · Codex
INSTANCE
your workspace
ACCOUNT
your own sign-in
SCOPES
named before granting
Scoped · revocable · tied to your own account
05 · Blast radius

Limits that hold when nobody is watching.

The question a security reviewer asks about agents is not whether they can be trusted, it is how much damage an untrusted one could do before somebody noticed. Every agent runs inside a spend ceiling, a token ceiling and a task rule, and its output does not reach main without a human gate. Automated review starts in record-only mode, so it writes down a verdict and moves nothing until you decide otherwise.

  • Monthly spend and token budgets per agent, plus expiring API keys.
  • One active task per assignee in each project by default, with stale-task detection for work that stops moving.
  • Human approval gates before agent work reaches main, with the rejection reason carried into the next attempt.
  • Record only is the default for automated review, so you can watch it agree with you before you let it move anything.
GUARDRAILS · PER AGENT
A runaway agent is a bounded problem.
Spend budget
per agent
Token budget
per agent
API key
expiring · rate limited
Active tasks
one per project
Merge
human gate
06 · Security scanning

A security gate you can enforce.

Every review comes with a security scan for committed secrets, vulnerable dependencies and risky code patterns. It runs on your own Station, where the code already is, so nothing has to be uploaded to be checked. Turn the gate on, and a change with a new critical or high finding cannot be approved or merged until the finding is fixed or deliberately waived.

  • Only new findings count. Issues already in your code do not block new work.
  • A waiver needs a reason, and records who granted it and which finding it covers.
  • Choose per project which scanners judge a change, including the ones you already run in CI. Checks for Azure Bicep and for every new npm and PyPI dependency.
  • Secret values stay on the machine. What reaches the server is the finding and where it is.
REVIEW REQUESTED · SCAN RUNS ON YOUR STATION
SECRETSCloud key in a test fixture
↳ value redacted before it leaves the machine
DEPENDENCIESLibrary with a known advisory
↳ introduced by this change · fixed version available
CODE PATTERNSNo findings
↳ against your rules
The finding leaves the machine. The credential does not.
Illustrative example, not a live deployment
07 · Audit

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, for workspace administrators. Connector activity is recorded to the same standard, each record naming the tenant, the user and the client. People with access to a project can watch its live agent sessions directly rather than asking for a screenshot.

  • System audit log with timeline view for workspace administrators, and audit export for compliance reviews.
  • Security evidence export for one task or a date range, as JSON, SARIF and HTML, with an SBOM for every scanned commit and whether the gate was on at each approval.
  • Connection events record declined connections as well as approved ones.
  • The live console follows project access, and streams the real session rather than a summary of it.
AUDIT TRAIL
Each record names the tenant, the user, and the client.
▸connector registered
▸connection approved
▸connection declined
▸access issued
▸access renewed
▸access revoked
Task moves, agent actions, approvals and policy gates are logged the same way.
08 · Isolation

One workspace cannot reach into another.

On a shared instance each workspace is kept to itself, so access to one workspace is never a way into another. Dedicated single-tenant instances are available where your review calls for one. On the workstation, every task runs in its own isolated workspace, so parallel work never shares a checkout.

  • Each workspace is kept separate on shared instances. Dedicated instances are single tenant.
  • Every task runs in its own isolated workspace on your Station.
  • Project environment documentation covers references and non-secret configuration. Secret values are not stored there.
ISOLATION
Shared instance
each workspace kept to itself
Dedicated instance
available, single tenant
Every task
its own isolated workspace on your Station
Environment docs
references and non-secret config only
Secret values are never stored in project documentation.
09 · Where things sit

Your code stays where your code already is.

Station runs on each operator’s own workstation and spawns the agent there, so source code is worked on inside your environment rather than uploaded to ours. What reaches Server is the record of the work: tasks, verdicts, decisions, knowledge and audit. Server runs on a production-grade database, and where it runs is your decision: most teams let us operate it, and it also runs on your own infrastructure. The air-gapped deployment requires that, because nothing on it reaches us at all.

  • Station runs Claude Code and Codex on your workstations, each task in its own isolated workspace.
  • Server holds the record of the work, behind your SSO, scoped by access levels.
  • Deployment is your choice: managed by Meshly, or self-hosted on your own infrastructure.
  • Air-gapped deployments are self-hosted by definition. Meshly AI runs local models on the box, and nothing leaves it.
WHERE THINGS SIT
Your code
Stays on your workstations. Station runs the agent there.
The record
Server, on a production-grade database. We host it, or you do.
Fully on-prem
Meshly AI runs local models on your own hardware.
10 · What we do not claim

The absent controls, listed by us rather than found by you.

Meshly Build has been through two security audits, each followed by hardening across the API surface. It has not been through a SOC 2 audit and it is not ISO 27001 certified, and we would rather you read that here than discover it three weeks into an evaluation. We are a young company operating a production system, and the honest description of our posture is engineering practice plus external review, not a certificate.

  • Two security audits, with the resulting hardening shipped in Server 2.0.0.
  • No SOC 2 report. No ISO 27001 certificate. We will say so before you ask.
  • If a specific control matters to your review, ask and you will get a direct answer about whether it exists today.
WHAT WE HOLD TODAY
Two security audits
Both were followed by hardening across the API surface, shipped in Server 2.0.0.
No certification claims
We hold no SOC 2 report and no ISO 27001 certificate. If a control matters to your review, ask us and you will get a direct answer about whether it exists.

Send us your questionnaire.

If your security review has a form, send it. You will get answers from the people who built the thing, marked clearly where the answer is no. If it raises a control we do not have yet, that is useful to us too.