Security built into every layer
Three-layer AI governance, tamper-proof audit trails, cryptographic agent identity, and bring-your-own AI model — built for teams that can't afford to get security wrong.
Every tool call passes three gates
Before any agent action reaches your systems, it must pass through three independent security layers. No single point of failure.
requests
tool call
Tool Classification
Every tool is classified as safe, restricted, or forbidden. Agents only access tools explicitly granted to them.
Tool Guard
Instant rule checks evaluate parameters before execution. Block or flag based on values, ranges, and business conditions. Zero latency.
Policy Shield
AI evaluates natural-language business rules for nuanced decisions. Fail-secure — if evaluation fails, the call is blocked.
tool call
executes
Sensitive actions pause
and wait for your approval
When an agent hits a sensitive operation, execution pauses mid-workflow. Your team sees the full context — what the agent wants to do, with what data, and why — then approves or denies with one click.
7 Approval Categories
Agent changes, tool access, mission execution, data operations, role changes, org settings, and tool data approval — each with its own policy.
Multi-Approver & Escalation
Require multiple sign-offs for high-risk actions. Set timeout rules that escalate to managers if nobody responds.
Tamper-Proof Execution
What was approved is exactly what executes. The payload is hashed at approval time and verified before execution — no modifications in between.
Every action. Timestamped. Tamper-proof.
Full decision traces for compliance reporting. Search, filter, and export millions of events — ready for your auditors.
Full Coverage
Auth events, agent operations, integration changes, knowledge updates, approvals, role changes, and every settings modification.
Powerful Search
Search millions of events by user, category, date, or resource. Export filtered results for external auditors in one click.
Execution Replay
Every mission run produces a step-by-step trace — tool calls, inputs, outputs, timing, and cost — replayable at any time.
Bring Your Own
AI Model
Keep complete control over your AI infrastructure. Connect models running in your own cloud — Azure OpenAI, AWS Bedrock, Google Vertex AI, or self-hosted endpoints. Your data never leaves your security perimeter.
Private Model Deployment
Connect to models in your own cloud tenant. No data passes through third-party AI providers unless you configure it.
Model Flexibility
Use different models for different agents. Pick the right balance of performance, cost, and compliance for each use case.
Data Sovereignty
Your operational data stays within your control. Choose your region, your provider, your encryption — Bixie adapts.
One Config Switch
Swap providers with a single configuration change. No code modifications, no mission rewrites, no downtime.
Fine-grained control over
who can do what
Single sign-on, 5 system roles, custom roles, and 34 granular permissions. Every user sees exactly what they should — nothing more.
| Capability | Owner | Admin | Editor | Member | Viewer |
|---|---|---|---|---|---|
| Create & edit missions | — | — | |||
| Run missions | — | ||||
| Approve actions | — | — | |||
| Manage integrations | — | — | — | ||
| Upload knowledge | — | ||||
| View audit logs | — | — | |||
| Manage team members | — | — | — | ||
| Org settings & billing | — | — | — | — |
Single Sign-On
Sign in via Okta, Azure AD, or any enterprise identity provider. Domain-based auto-join and enforced SSO.
Multi-Tenant Isolation
Every query is scoped to the organisation. Data isolation is enforced at the infrastructure level, not application-level filtering.
Emergency Access
Break-glass access for critical situations with full audit logging. Time-limited, requires owner approval.
Cryptographic agent identity.
Zero implicit trust.
Every agent proves its identity before it can call a single tool. Environment verification, scoped access tokens, and an audited tool gateway — your systems never trust an unverified caller.
Trust Levels
Why This Matters
No network-level trust
Even if an attacker is inside your network, they can't call tools without a valid environment proof.
Short-lived by design
Tokens expire quickly. A compromised token has a narrow window before it's useless.
Real-time risk signals
Unusual request patterns, new agents, or anomalous timing automatically trigger stricter checks.
Hash-chained audit
Every record is cryptographically linked. Tampering with any entry breaks the chain and raises alerts.
Encrypted everywhere. Always.
Independently verified.
Continuously audited.
Our security practices are validated by independent third-party auditors against the toughest industry standards.
Security, availability, and confidentiality — audited annually
International standard for information security management
Operational Security
Found a vulnerability?
We take every report seriously and respond promptly. Responsible disclosure is appreciated — we'll work with you to understand and resolve any issue.
All Bixie products and services are subject to our comprehensive security controls, policies, and certifications. For questions about our security practices, contact support@team.bixie.ai.