Inspiration: Frustration with opaque agent loops and security/compliance gaps; desire for transparent, auditable, enterprise-safe agent runtime.
What it does: Defines agents as declarative YAML, wraps execution in security/compliance layers, enforces schema validation, GDPR/LGPD checks, threat detection, immutable audit trails, and sandboxed runtime with Gemini helpers.
How I built it: Hardened FAILSAFE YAML parser; declarative interpreter + Zod schema compiler; isolated-vm sandbox manager; compliance engines (GDPR/LGPD); threat detection and audit services; performance monitors and autoscaling helpers; Firebase multi-tenant orchestration; packaged in beddel with example Next.js app.
Challenges I ran into: Avoiding arbitrary code execution while allowing flexible agents; enforcing strict schema validation without hurting DX; isolating runtime resources per tenant; balancing security scans with performance; handling PII/consent automatically.
Accomplishments that I'm proud of: Declarative protocol for portable agents; deterministic security guardrails (no eval); built-in compliance engines and audit hashing; Gemini Flash helpers with strict validation; multi-tenant isolation with runtime profiles.
What I learned: Declarative schemas make agent behavior auditable; security/compliance must be first-class in agent design; sandboxing + validation layers are essential to prevent code-execution risks from generative systems.
What's next for beddel: Expand declarative language (pipelines, state machines); richer behavior registry/marketplace; more integrations and orchestration controls; deeper compliance automation; community feedback on schema and enforcement approach.
Built With
- firebase
- genkit
- node.js
- typescript
- yaml