Observe, Guard & Self‑Heal Your AI Agent Actions

Staso watches & evaluates every agent action, blocks mistakes before they execute, and fixes the patterns that cause them. One platform. Three lines of code.

PII redacted before storage
The Problem

AI agents fail in production. Every day.

Unauthorized actions

Agent executes high-impact operations without human approval — refunds, deletions, account changes.

Runaway loops

Agent retries the same failing call dozens of times. You find out from your cloud bill.

Hallucinated completions

Agent tells a user their issue is resolved when it isn’t. Fabricates data mid-conversation.

Leaked credentials

Agent logs an API key, a customer’s SSN, or a session token into your trace data. Now it’s in your observability vendor’s database. Forever.

Observability shows you what happened. That's not enough. You need infrastructure that stops it from happening.

The Platform

From visibility to prevention to healing

Observe every agent decision. Block the dangerous ones. Fix the patterns that cause them — automatically.

Observe

Execution Traces

Every tool call, LLM request, and agent step — captured with full context. You can't fix what you can't see.

Monitoring Dashboard

Error rate, P95 latency, token usage, and cost in real time. Know when something breaks before your users do.

Conversations & Cost Tracking

Group traces into multi-turn conversations. Track spend per trace, agent, and model.

Datasets & Eval Runner

Curate real traces into test data with one click. Excel-like editor, splits, CSV import/export, and an eval runner that scores every row against your own scorers.

Built-in PII Redaction

SSNs, credit cards, API keys, and secrets are stripped from traces before they hit storage. On by default. Zero config.

TracesSpan treeWaterfallDashboardConversationsCost trackingDatasetsEval runnerPII redactionSecret scrubbing

Runtime Firewall

Live

Static + LLM Rule Engine

19 zero-config rules detect PII, secrets, injections, and dangerous operations instantly. 8 LLM judges catch what regex can't — wrong tools, hallucinated args, jailbreaks, and cost escalation.

Custom Rules & Policies

Write rules in plain English. Group them into policies. Scope by environment, agent, or tool. Audit first, enforce when ready — per rule, per environment.

Evaluation Context Scope

Evaluate at tool-call level for granular control, or at full trace level for end-to-end validation. Choose the scope that fits each rule.

Evidence-Backed Validation

Every agent decision validated against an evidence cache of prior tool call outputs. Blocks actions based on hallucinated data before they execute.

19 static rules8 LLM judgesCustom rulesAudit / EnforceTool-level scopeTrace-level scopeEvidence cacheRBAC

Evaluate

Coming Soon

Input Scanning

Block prompt injection, jailbreaks, and credential exposure before they reach your agent.

Output Validation

Flag hallucinations, false completion claims, quality drift, and reasoning errors — at the tool-call level, where the expensive mistakes happen.

Injection detectionHallucination detectionEvidence cache

Self-Heal

Live

Code-aware diagnosis

Staso reads your repo at the exact version your agent was running and traces failures to their root — the commit, the function, the pattern that broke.

Fix PRs

Turn any diagnosis into a pull request. Staso writes the fix, verifies it in a sandbox, and opens a PR. Review the diff, merge when ready.

Code-aware diagnosisFix PRsSandbox deep-diveCheckpoint replaysoonRegression detectionsoon
Why Staso

Built differently

01

One system that actually learns

Firewall catches the pattern. Traces record what happened. Auto-fix diagnoses root cause and updates the detection rule. Every layer shares the same data and compounds on itself.

02

Gets smarter per customer

Every blocked action and detected failure feeds a pattern database specific to your agents. Detection sharpens over time because the system learns what “bad” looks like for your use cases.

03

Works the moment you install

Ships with proprietary detection rules for prompt injection, PII leakage, jailbreaks, and dangerous operations. No configuration. No empty dashboards. Protection from minute one.

04

Three lines of code

pip install staso. Add three lines. Done. Built for developers who ship fast, not procurement teams who evaluate for months.

How It Works

Ship in under 5 minutes.

Wrap your existing client or run one CLI command. No agent rewrites. No config files. No YAML. Works with OpenAI, Anthropic, Claude Code, and Codex today.

Every tool call renders as a human-readable card on your dashboard -- diffs, terminal output, agent trees, plan steps. Not raw JSON.

Read full documentation
Works with
# Install
$ pip install staso

# Done. Every session traced.
$ staso setup
Early Adopters

What developers are saying

“Honestly just needed something to see what my Claude Code agent was doing across runs. Staso was the quickest to set up and the trace view is clean.”

Jashan Sandhu

Senior Software Engineer

“I run almost everything through OpenClaw these days. The observability part in Staso is decent for watching what my agent does across sessions. Waiting for the enforcement layer though — that’s the part I actually need.”

Ishant Gupta

Full-Stack Developer

“Asked the founders if they could support Codex and they shipped it within a day. Using it to monitor my Codex runs now — minimal, shows what I need, nothing extra.”

Ramit Shivansh

Senior Software Engineer

“Honestly just needed something to see what my Claude Code agent was doing across runs. Staso was the quickest to set up and the trace view is clean.”

Jashan Sandhu

Senior Software Engineer

“I run almost everything through OpenClaw these days. The observability part in Staso is decent for watching what my agent does across sessions. Waiting for the enforcement layer though — that’s the part I actually need.”

Ishant Gupta

Full-Stack Developer

“Asked the founders if they could support Codex and they shipped it within a day. Using it to monitor my Codex runs now — minimal, shows what I need, nothing extra.”

Ramit Shivansh

Senior Software Engineer

Pricing

Start free. Scale when you're ready.

10,000 traces/month free. No credit card required. Upgrade when your agents hit production volume.

Personal

Full observability for individual developers. Free forever.

Free
  • 10,000 traces / month
  • 7 days data retention
  • 3 workspaces
  • Up to 3 team members
  • 5 custom rules & policies
  • 10,000 static guard evaluations / month
  • 300 LLM guard evaluations / month
  • 3 evaluation datasets
Start Free

Team

Popular

Production agents need production infrastructure.

Coming Soon

Everything in Personal, plus:

  • 100,000 traces / month
  • 30 days data retention
  • 20 workspaces
  • Up to 10 team members
  • 10 custom roles (RBAC)
  • 30 custom rules & policies
  • Unlimited static guard evaluations
  • 5,000 LLM guard evaluations / month
  • 30 evaluation datasets
  • Priority support

Enterprise

Your agents, your rules. Everything unlimited.

Custom

Everything in Team, plus:

  • Unlimited traces
  • Custom data retention
  • Unlimited team members
  • Unlimited API keys & workspaces
  • Unlimited custom roles & SSO
  • Unlimited guard evals
  • Unlimited rules & policies
  • Unlimited datasets
  • Dedicated support & SLA
Book a Demo

See Staso on your agents

15-minute walkthrough. We connect to your stack, show you live traces, and demo the runtime firewall on your actual agent traffic.

FAQ

Frequently Asked Questions

Observability tells you what happened after the fact. Staso intercepts agent decisions before they execute. Observe, block, evaluate, and fix — in one platform sharing the same data.