Show HN: Samma Suit – Open-source 8-layer security framework for AI agents
sammasuit.comI've been running AI agents in production for a music platform and kept hitting the same security gaps — no permission inheritance, no cost controls on nested calls, skills that could execute arbitrary code, no kill switch when things went sideways. Samma Suit is an open-source security layer that wraps any agent framework with 8 layers:
SUTRA — API gateway with rate limiting DHARMA — Permission inheritance (parent → child agents) SANGHA — Skill/tool vetting before execution KARMA — Cost controls that propagate through subagents SILA — Immutable audit trail METTA — Cryptographic identity signing BODHI — Execution isolation NIRVANA — Kill switch for runaway agents
Framework-agnostic — works with LangChain, CrewAI, AutoGPT, or raw API calls. Define policies in YAML, Samma Suit enforces them at runtime.
GitHub: https://github.com/onezeroeight/samma-suit
Docs: https://sammasuit.com
The arxiv paper on the front page today (agents violating constraints 30-50% of the time) is exactly why we built this — constraints need to be enforced at infrastructure level, not left to the model. A bit more on the architecture:
Each layer is a middleware that wraps the agent's execution loop. When an agent calls a tool or spawns a subagent, the request passes through the stack:
Agent Request → SUTRA (rate limit) → DHARMA (permissions) → SANGHA (skill check) → KARMA (cost) → Execute
↓
Agent Response ← SILA (audit log) ← METTA (sign) ← BODHI (isolate) ← NIRVANA (kill if needed) ←
Policies are YAML:
yamlpermissions:
file_system:
read: ["/data/*"]
write: []
network:
allowed_domains: ["api.anthropic.com"]
cost:
max_per_request: 0.10
max_per_session: 5.00 kill_conditions:
- token_count > 100000
- execution_time > 300s
- error_rate > 0.5
The key insight from running agents in production: most failures aren't the model being malicious — they're the model being helpful in ways you didn't anticipate. DHARMA and SANGHA catch those before they execute.
Happy to go deeper on any layer.