Egress Allowlists & Audit Streams
AgentSecrets Cloud combines an egress domain firewall with a tamper-evident audit stream, providing containment and complete observability over all credential usage.
Egress Domain Allowlist Firewall
Autonomous AI agents executing multi-step tasks must be bounded to authorized third-party APIs. The Egress Allowlist prevents malicious prompts, hallucinatory outputs, or compromised dependencies from exfiltrating credentials to unauthorized endpoints.
Allowlist Evaluation Rules
- Exact Domain Match:
api.stripe.compermits requests exclusively tohttps://api.stripe.com/*. - Wildcard Subdomain Match:
*.openai.compermitsapi.openai.comandv1.api.openai.com, while rejectingunauthorized-openai.com. - Port Stripping: Automatically normalizes hostnames across ports (
api.github.com:443->api.github.com). - SSRF & Private IP Blocking: Outbound calls to non-routable private IP addresses (
127.0.0.1,10.0.0.0/8,169.254.169.254,172.16.0.0/12,192.168.0.0/16) are blocked unconditionally to protect cloud instance metadata services.
Enforcement Behavior on Blocked Requests
If an agent attempts to resolve a credential for a non-allowlisted destination (such as https://attacker-c2.com/collect):
- The Cloud Resolver terminates the connection immediately with
403 Forbidden(domain_blocked). - An audit record is logged flagged with the blocked domain and timestamp.
- The credential value is never decrypted, unsealed, or transmitted.
Live Audit Streams
Every credential resolution—whether executed through agentsecrets env container boots or POST /v1/resolve agent tool calls—emits a structured, zero-knowledge audit event.
AUDIT INGESTION PIPELINE ┌──────────────┐ │ HTTP Request │ └──────┬───────┘ ▼ ┌─────────────────────────────┐ │ Cloud Resolver Data Plane │ │ (Processes & Logs Entry) │ └──────────────┬──────────────┘ │ Lock-Free Push ▼ ┌─────────────────────────────┐ │ Bounded Ring Buffer │ │ (10,000 Capacity) │ └──────────────┬──────────────┘ │ Batched Flush (500ms or 100 items) ▼ ┌─────────────────────────────┐ │ PostgreSQL Audit Database │ │ & Real-Time Web Stream │ └─────────────────────────────┘
Audit Record Schema
Audit events capture complete execution telemetry without storing credential values:
| Field | Type | Description | Example |
|---|---|---|---|
timestamp | ISO-8601 | High-precision UTC timestamp. | 2026-08-23T00:15:02.124Z |
workspace_id | UUID | Workspace owning the credential. | 8fa3c012-7b89-... |
project_id | UUID | Project context. | 1ab4d982-3c4e-... |
agent_token_id | string | Workload token ID used. | agt_prod_7f8a... |
credential_key | string | Secret handle referenced (name only). | STRIPE_KEY |
target_domain | string | Destination hostname. | api.stripe.com |
target_path | string | HTTP path requested. | /v1/charges |
method | string | HTTP method. | POST |
status_code | integer | Upstream HTTP response status code. | 200 |
duration_ms | integer | Round-trip latency in milliseconds. | 142 |
redacted | boolean | true if reflected credentials were scrubbed. | false |
resolution_path | string | direct_resolve or headless_env. | direct_resolve |
Observation & Management Interfaces
1Web Dashboard Audit Viewer
Navigate to Audit Logs in the AgentSecrets Web App for live log streaming, domain filtering, error inspection, and CSV export.
2CLI Inspection Commands
# View recent audit events agentsecrets logs --limit 50 # Filter audit records by target domain agentsecrets logs --domain api.stripe.com # Stream incoming audit log events in real time agentsecrets logs watch # List active domain allowlist rules agentsecrets allowlist list # Add new domains to the workspace allowlist agentsecrets allowlist add api.stripe.com api.github.com