Pre-action control
Risky actions pause before they reach customers or business systems.
TRUST & SECURITY
Trust starts before the agent acts.
Argos is built around a simple principle: risky agent actions should be reviewed before execution, and every decision should leave a clear evidence trail.
Minimum data by default
Argos is designed around metadata and redacted summaries, not raw secrets or full records.
Evidence trail
Approval decisions, risk reasons, and final outcomes are recorded for review.
Data boundaries
Argos should receive enough context to evaluate an action, not unnecessary raw business data.
What Argos receives
- Agent ID
- Workflow ID
- Tool and action metadata
- Target type
- Business object summary
- Redacted payload summary
- Policy decision
- Reviewer decision
- Final outcome
What Argos does not need by default
- Raw email bodies
- Customer lists
- API keys
- Credentials
- Full CRM records
- Private prompts
- Secrets
Teams should send the minimum context required to evaluate and approve the action.
How control works
A security story around the action, not just another log after it happens.
The control path is designed to make risk, approval, and evidence explicit before a sensitive action reaches a customer or business system.
- 1
Action attempted
An agent tries to send, update, trigger, delete, or modify something.
- 2
Policy checked
Argos evaluates tool, action, target, payload summary, and business context.
- 3
Approval routed
High-risk actions are sent to a human reviewer before execution.
- 4
Decision recorded
Approve, deny, block, or escalate decisions are stored with context.
- 5
Outcome saved
The final result updates the evidence packet.
What Argos stores
Evidence around the decision, not full raw business systems by default.
Argos stores the approval and evidence record around the action, not full raw business systems by default.
- Action attempt metadata
- Risk explanation
- Approval request
- Reviewer decision
- Evidence packet
- Timeline
- Policy snapshot
- Final outcome
Access control
Argos separates agent API access from human reviewer access.
Agent calls and reviewer decisions use different trust paths, so approval remains a human-controlled step.
Agent/API access
- API keys are scoped to project and agent
- API keys are revealed once and stored as hashes
- Revoked keys cannot authenticate
Reviewer access
- Reviewers approve through authenticated app access
- App access is scoped through trusted memberships
Evidence packets
A reviewable record for the action that almost happened.
Evidence packets are designed to show what action was attempted, why it was risky, who approved or denied it, and what happened afterward.
Evidence packet
Stopped before execution
- 1
AI drafted
- 2
Argos paused
- 3
Human approved
Before executionHuman reviewEvidence saved
Email sent after review
Deployment and data posture
Metadata-first by default, with deployment needs discussed during the pilot.
Argos starts with metadata and redacted summaries by default. Regional, private, and self-hosted deployment requirements can be discussed with design partners.
- Metadata-first architecture
- Redacted summaries by default
- Private deployment available on request
- Self-hosted deployment available on request
- Retention controls are part of the roadmap
Argos is currently in private beta. We are working with design partners to harden security, workflows, and deployment requirements before broader release.
Compliance note
Argos is not a substitute for legal, regulatory, or compliance review. It helps teams create approval records, policy snapshots, reviewer decisions, and action evidence that can support internal governance and audit workflows.
Want to review your riskiest agent workflow?
Apply for the Design Partner Pilot and we will help map where your first approval checkpoint should sit.