Agent Serve · Security
Single sign-on, self-hosting in your own cloud and zero data retention with supported model providers, backed by the controls your auditors ask for.
Agent Serve, the AI workspace where teams build, deploy, and manage AI agents, is built for teams running SOC 2 and GDPR programs. It provides SAML and OIDC single sign-on with SCIM provisioning, self-hosting with a Helm chart or Docker, zero data retention with supported model providers, an exportable activity audit, access policies, row-level security, approvals, data retention settings, PII redaction, log destinations, encryption and invite-only access. Ready means the controls are in place; it is not a certification.
Out of the box
Agent Serve supports SAML and OIDC single sign-on, with SCIM 2.0 to provision people and groups automatically.
Deploy Agent Serve inside your own cloud account or VPC, next to your data, with the same product your team sees here.
Run models under your organization’s zero data retention terms, or keep inference inside your network, and decide how long Agent Serve keeps logs.
SOC 2 ready and GDPR ready mean the controls are in place. Agent Serve does not claim a SOC 2 report or a certification; certification is your organization’s or a future audit’s scope. Zero data retention is backed by your own provider keys and agreements, self-hosted models, configurable log retention and PII redaction. Some provider features keep data on the provider’s side by design — for example Gemini Deep Research requires stored interactions — so they’re outside zero data retention.
In depth
Each Agent Serve control lives under Organization & governance in settings, where admins set it and auditors can see it.
Agent Serve supports SAML and OIDC single sign-on with verified domains, and SCIM 2.0 provisions people and groups from your identity provider.
Agent Serve Access policies limit which blocks, models and integrations a permission group can use, on top of admin, write and read roles.
The Agent Serve Activity audit lists workspace and organization activity, including Relay approvals and declines, and exports it as CSV.
Agent Serve keeps data from 1 day to 5 years, or forever, and redacts 40 types of personal data at the Logs, Workflow input and Block outputs stages.
Agent Serve log destinations send workflow logs, job logs, audit logs and Chat history to your storage or monitoring tools every hour or every day.
By the numbers
7
Log destinations
40
Personal data types detected
3
Stages where redaction applies
3
Roles per team space: admin, write, read
Counts are read from the product code.
Onboarding
How an organization brings people into Agent Serve, with identity, access and audit set before the first agent runs.
Verify your domain and connect SAML or OIDC single sign-on.
SCIM 2.0 creates people and groups from your directory and removes leavers.
Permission groups decide which blocks, models and integrations each group can use.
The Activity audit records each change, and log destinations stream records to your systems.
Supporting controls
Each control below is a setting or feature in Agent Serve today. Some are Enterprise features.
Agent Serve Relays pause before external actions, and workflows can wait for a person at any step.
Send email → Needs approval
Agent Serve exports logs, tables, knowledge, files, dashboards and workflows, and people can delete their own account.
Per-resource export
Agent Serve encrypts secrets and credentials with AES-256-GCM. Reference cloud deployments encrypt storage at rest and use TLS in transit.
AES-256-GCM · TLS 1.2+
Agent Serve accounts are created by invitation. Session lifetime and idle timeouts are set per organization.
Invite → SSO → Session policy
The reference Agent Serve AWS deployment runs in eu-central-1 (Frankfurt) by default, or in the region you choose when you self-host.
eu-central-1
Agent Serve uses your organization’s own model provider keys, set at organization or team space level.
Settings → Model API keys
Every Agent Serve workflow run records each block’s inputs, outputs and errors.
Run → Block → Output
Agent Serve Guardrails check model output for valid structure, patterns, unsupported claims or personal data before it moves on.
Hallucination check
Agent Serve ships a Helm chart, Docker images with a production compose file, and reference Terraform for AWS.
Helm · Docker · Terraform
Before and after
| Work | Before | With Agent Serve |
|---|---|---|
| Who can sign in | BeforeSeparate passwords, and accounts left behind when people leave. | With Agent ServeAgent Serve single sign-on with SCIM, so leavers lose access with your directory. |
| What AI agents may do | BeforeAn assistant acts with whatever access its key has. | With Agent ServeAccess policies, row-level security and approvals before external actions. |
| Audit evidence | BeforeScreenshots collected by hand before each review. | With Agent ServeAn Activity audit exported as CSV, or streamed to your storage on a schedule. |
| Personal data in logs | BeforeKept forever, unredacted, in whatever the tool stores. | With Agent ServeRedacted at three stages and kept only as long as your retention setting. |
Under the hood
| Single sign-on | SAML 2.0 and OIDC with verified domains (Enterprise). SCIM 2.0 provisioning. |
|---|---|
| Roles | Admin, write and read per team space, plus permission groups. |
| Access policies | Limit the blocks, models and integrations each permission group can use. |
| Sessions | Maximum session lifetime, idle timeout, and Sign out all members. |
| Activity audit | Organization and workspace activity, exported as CSV. |
| Data retention | 1 day to 5 years, or forever, per organization or team space (Enterprise). |
| PII redaction | 40 personal data types at the Logs, Workflow input and Block outputs stages, plus your own patterns. |
| Log destinations | Amazon S3, Google Cloud Storage, Azure Blob Storage, Datadog, Google BigQuery, Snowflake or an HTTPS webhook, every hour or every day. |
| API keys | Team space and personal keys. Admins can turn off personal API keys. |
Works with
Relays ask for approval before email and meetings, and every decision is audited.
Redact personal data in workflow input and block outputs, and trace every run.
Row-level security, locked filters and expiring viewer links on shared dashboards.
Row-level security and column restrictions on workspace tables.
Use your own model provider keys, or self-hosted models inside your network.
Legal
How Agent Serve handles personal data, the terms of use, and who processes data for us.
Further reading
A regulator’s letter becomes a dozen questions, a dozen owners and one deadline. Here is a setup where a workflow logs and splits the request, a Relay drafts from approved sources and a named person signs off every answer.
Regulatory affairs · 13 min readEvery week the committee secretary rebuilds the same pack from memos, spreadsheets and emails. Here is a setup where four workflows assemble it from the source systems and the secretary decides when it goes out.
Commercial lending · 12 min readMost security questionnaires ask the same two hundred questions in a different order. Here is a workflow that drafts each answer from your own policies and past approved answers, and sends anything uncertain to the right expert.
Sales and security · 13 min readQuestions
No. Agent Serve is SOC 2 ready: the controls auditors look for, such as SSO, audit logs, access control, retention and encryption, are in place. Agent Serve does not hold a SOC 2 report today.
Agent Serve is GDPR ready. It gives teams the controls a GDPR program relies on, including regional hosting, retention settings, PII redaction, data export and account deletion. Compliance depends on how your organization uses it.
As long as your Agent Serve data retention setting says: from 1 day to 5 years, or forever, set per organization or team space.
Agent Serve works with model providers that offer zero data retention. It is backed by your organization’s own provider keys and agreements, self-hosted models that keep prompts in your network, configurable log retention and PII redaction. Some provider features keep data on the provider’s side by design — for example Gemini Deep Research requires stored interactions — so they’re outside zero data retention.
No. Valnora & Elric does not sell personal data and does not use your data to train AI models. To keep prompts out of a model provider’s systems, use zero data retention with a supported provider, your own provider keys, or a self-hosted model.
Yes. Valnora & Elric offers a Data Processing Agreement (DPA) for Agent Serve customers. Email systems@valnoraelric.com to request one.
Yes. Agent Serve ships a Helm chart for Kubernetes, Docker images with a production compose file, and reference Terraform for AWS, so it can run in your own cloud or VPC.
Last updated
Enterprise access
Agent Serve is available to teams by invitation. Tell us what you want to build and we will set up your workspace with you.