ABOUT · STASO AI

Infrastructure for agents in production.

Guard. Evaluate. Observe. Use Staso Agent | The loop every team running agents ends up building or skip entirely — shipped as one platform.

01 · WHY WE'RE BUILDING THIS

We felt the maintenance loop.

“We have used agent tools at our jobs, and felt the constant need for a maintenance loop once an agent hits production.”

Existing platforms specialise — one for tracing, one for evals, one for post-mortem RCA. Stitching them costs context at every boundary, and most teams use less than ten percent of any one of them.

Staso is the opposite bet: runtime firewall, evaluations, observability, and Staso Agent under one roof, sharing one data model. The system gets smarter per customer because the full loop lives in one place.

02 · TEAM

Two founders. Shipping fast.

Mayank Sheoran
Co-founder · CEO

Mayank Sheoran

Built production systems across enterprise SaaS, e-commerce & fintech

Zomato · Navi · CodeAnt AI (YC24) · Fampay (YC19)

linkedin →
***
Co-founder · CTO

***

***

***

linkedin →
03 · WHERE WE ARE

Four pillars. One data model.

Everything below is in production today. The roadmap under each pillar is where it sharpens next.

Runtime firewall

Block it before it runs.

Live
  • 30+ Zero-config rules — PII, secrets, prompt injection, jailbreaks ...
  • Custom LLM-judge
  • Policies with audit or enforce mode
  • Force-audit override per rule
Soon
  • More zero-config rules & guardrail marketplace.
Evaluations

Regress what you just fixed.

Live
  • Curate datasets from traces or with your own data
  • Splits, CSV import / export, spreadsheet design
  • SDK-side evaluate() with auto trace capture
  • Smart column extractors on curate
  • Server-side zero-config and custom eval rules
  • LLM-judge and Agent Judge grading for complex data
  • Promote an eval rule into a real-time guard block
Soon
  • Synthetic row generation
Staso Agent

From question to tested pull request.

Live
  • Long-lived coding conversation
  • Trace, span, guard, eval, and verdict tools
  • Sandboxed repository edits and tests
  • Pull requests from the tested workspace
Observability

Every decision, visible.

Live
  • Traces, spans, live WebSocket stream
  • Conversations stitched across sibling LLM calls
  • Pretty engine for Claude, OpenAI, Codex payloads
  • Cost, tokens, agent metrics, environment filters
Soon
  • Threshold alerting on latency, cost, error rate

Hosting options, SOC 2, HIPAA on the platform roadmap.

04 · Talk to us

Building with agents? We want the hard cases.

Design partners get direct founder access and first look at every new rule.

Start tracing

or email [email protected]