Portfolio • Book a Call • LinkedIn • X • Email
I'm a Tech Lead at Attri.ai. By day I build production multi-agent systems. After hours I build the open-source tools that police them: agent-audit-kit (a static scanner for MCP and agent pipelines), agent-airlock (a deny-by-default contract layer for agent tool calls) and Provael (red-teaming robot policies). I publish the numbers, including the nulls.
Highlights
- The State of MCP Security 2026: I scanned 2,303 public MCP server configs; 50.6% declare a remote server with no authentication. Reproducible offline. Read the report
- Provael: red-team and assurance for vision-language-action robot policies, in simulation, with every result reported against a benign control and a 95% interval. Black Hat Europe 2026 Arsenal. I also wrote the Embodied AI Security Top 10. provael.com
- Agentify: Attri.ai's flagship multi-agent platform (15+ specialized agents), with a cascading model router and E2B micro-VM sandboxing
- MannSetu: solo-built, voice-first AI mental wellness companion for India, live at mannsetu.com with early users
- pyAGI: an early open-source Python agent framework (2023), acquired by Kyle Morris (co-founder of Banana.dev) and Jeffrey for $10,000
10-second summary: I ship production agents. And police them.
| Domain | Details |
|---|---|
| Multi-Agent Systems | Role design, tool calling, memory, governance, deterministic workflows |
| GenAI Engineering | Cascading model routing, budget controls, evals, guardrails, RAG, micro-VM sandboxing |
| Agent Security | Static scanning, runtime firewalls, capability leases, OWASP Agentic Top-10, red-teaming |
| Cloud Architecture | AWS + Azure (primary), GCP, multi-tenant SaaS, IaC, queues, caching |
| Ops Mindset | Observability (OpenTelemetry, Datadog, Sentry), incident response, cost visibility, SLAs |
I'm building the open-source security infrastructure for AI agents: static analysis, runtime firewalls, and operational control planes, now extending to embodied, physical AI.
| Layer | Project | What It Does |
|---|---|---|
| Pre-deployment scanning | agent-audit-kit | Static scanner for MCP and agent pipelines: 415 rules (v0.6.25), SARIF, GitHub Action, OWASP Agentic and MCP Top-10 coverage |
| Runtime firewall | agent-airlock | Deny-by-default contract layer for agent tool calls: argument validation, ghost-argument stripping, capability leases |
| Operations control plane | ferrumdeck | Rust governance engine, budget enforcement, deny-by-default policies |
| Agent memory | mnemo | On-prem, MCP-native agent memory with a hash-chained log an auditor can verify |
| Browser security | zerotrust | On-device AI threat detection, privacy-first, runs entirely in browser |
| Embodied / physical AI | Provael | Red-teams VLA robot policies in simulation; attack-success rate against a benign control with a 95% interval; Black Hat Europe 2026 Arsenal |
|
agent-audit-kit Static scanner for MCP-connected agent pipelines. 415 rules, OWASP Agentic and MCP Top-10, GitHub Action, SARIF, and the State of MCP Security 2026 census. 
|
agent-airlock Deny-by-default contract layer for agent tool calls: strict argument validation, ghost-argument stripping and capability leases, in-process. 
|
ferrumdeck Production-grade AgentOps control plane. Rust governance engine + Python LLM runtime + Next.js dashboard. Deny-by-default policies, budget enforcement & audit logging. 
|
|
mnemo On-prem, MCP-native memory for AI agents, in Rust. Every write lands in a SHA-256 hash-chained log an auditor can verify offline. 
|
zerotrust On-device AI Chrome extension for real-time website security analysis. Privacy-first threat detection that runs entirely in the browser; no data leaves your machine. 
|
provael Red-team and assurance for VLA robot policies, in simulation. Results against a benign control with a 95% interval. Black Hat Europe 2026 Arsenal. 
|
More Notable Projects
| Project | What it is | Links |
|---|---|---|
| pyAGI | Early open-source Python agent framework (2023); acquired by Kyle Morris and Jeffrey for $10,000 | Repo ·  |
| inksight | Forensic signature verification with explainable AI (pip install inksight) |
Repo ·  |
| AgentX | Unified agent framework playground (tools + eval + obs) | Repo ·  |
| PrimeAcrossAll | Same problem across many languages (Docker-based) | Repo ·  |
| Dynamic-ML-Orchestration | Serverless ML orchestration patterns | Repo ·  |
| PyVerseAI | Collection of AI + cloud projects | Repo ·  |
- Newsletter and book: The Production Agent, how to ship agentic AI to production without getting paged at 3am (chapters 1 and 2 drafted, free to read)
- Medium: medium.com/@sattyamjain96
- Dev.to: dev.to/sattyamjjain
- Website: sattyamjjain.in
I'm open to:
- Advisory + consulting (agent architecture, security, cost optimization)
- Open-source collaborations (especially agent governance & physical AI)
- Speaking at conferences and meetups on production AI systems
