workspace / mitwa-ai / overview

Search runs, envs, validators…⌘K

🔔

⚙

Dashboard

Good evening, Sumedh.

Tuesday, May 5 · 21 active episodes across 7 environments

3 runs streaming

Total Runs · 7D

1,284

▲ 18.2% vs prev

Avg Score

0.847/10

▲ 0.034

Pass Rate

91.4%

▼ 1.1%

Compute · 7D

412cpu·hr

▲ 22.4%

Recent Runs

Open all →

RUN IDENVIRONMENTTASKSCORESTATUSDURATION

r_8f3a21crm_v1t3_create_deal_workflow0.91PASS14.2s r_8f3a20browser_v2checkout_flow_30.67PARTIAL38.1s r_8f3a1fhealthcare_v1t4_phi_scope_safety1.00PASS22.7s r_8f3a1eterminal_v3git_rebase_conflict—RUNNING—r_8f3a1ditsm_v1t2_route_priority_sla0.21FAIL11.8s r_8f3a1ccrm_v1t1_lookup_email1.00PASS3.4s

Score Trend · 14D

your agentbaseline

14d delta+0.12 vs baseline· +14.7%↑

Environments

DB

crm_v1

Salesforce-style · 112 tasks · contacts, deals, pipelines

v1.2.4412 runs0.89—

{}

browser_v2

Playwright sandbox · 128 tasks · DOM observation

v2.0.1287 runs0.71—

+

healthcare_v1

Scheduling + EHR · 84 tasks · HIPAA-scope safety

v1.4.0198 runs0.92—

$_

terminal_v3

Sandboxed shell · 96 tasks · gVisor isolation

v3.1.0156 runs0.42—

#

itsm_v1

Ticket triage · 76 tasks · SLA routing

v1.0.7131 runs0.61—

$$

erp_v1

Procurement + AR/AP · 104 tasks · approval graphs

v1.3.262 runs0.85—

/>

api_mock_v2

REST schema · 68 tasks · stateful fixtures

v2.1.038 runs0.94—

Resource Usage

CPU

42%

Memory

68%

Storage

24%

Network

31%

Concurrent

21 / 100

Quota · runs

84 8.4k / 10k

Validator Coverage

7-day rolling

PROGRAMMATIC

94.2%

LLM JUDGE

81.6%

HUMAN SPOT

12.4%

FAILURE TAX.

38 classes

Reports

□

Weekly evaluation summary

All envs · pass rate, regressions, top failures

CRM v1.2.4 release report

Diff vs v1.2.3 · 412 runs replayed · safety pass

Browser_v2 regression analysis

checkout_flow_3 · root cause + replay

Healthcare HIPAA certification

v1.4.0 · 84/84 tasks · external auditor signed

PDF · 22p

updated 3d ago

→

Compute cost breakdown

April · per-env · per-tier

CSV

updated 4d ago

→

Quarterly benchmark vs baseline

Q1 2026 · 6 envs · vs gpt-4o, claude-3.5

HTML

updated 2w ago

→

Docs

QUICKSTART

Install & first run

Get from zero to a scored episode in under five minutes.

SDK

Python SDK reference

Connect any agent loop to a REST step endpoint.

CONCEPTS

Validators & scoring

How programmatic, judge, and human signals combine.

CI/CD

GitHub Action setup

Block PRs that regress your agent suite.

GUIDE

Writing a custom env

From a brief to schema, tasks, and validators.

API

REST endpoints

/episode, /step, /score, /reset — all documented.

RECIPES

Replay & debugging

Diff your agent against canonical replays.

ENTERPRISE

SSO, VPC, audit logs

Deploy in your network with full compliance.

CHANGELOG

v0.4 release notes

April 2026 — public beta, new env library, judges.

region: us-east-1 · build a1f3c2 · all systems normallast sync · just now