Basenull AI Ops · Multi-Agent Infrastructure

Make sure your AI agents
actually finish the job.

Process monitoring for AI agents. Define a workflow, watch your agent run it, get alerted when it skips a step or stops running on time.

Every team is shipping agents. Most pay attention the first few weeks, then move on. Months later the model drifts, an upstream API changes, and your workflow is silently broken — until something downstream visibly fails. agent-check catches the drift the day it happens, not three weeks later.

Free · 1 workflow · 100 verifier credits · no credit card

How it works

Define a workflow

Paste your agent's existing instruction prompt; we draft a list of tasks with expected evidence for each. Tweak, save.

Wire your agent

Drop the MCP config snippet into your agent's setup. The agent calls start_run, submits evidence per task, calls complete_run.

Get alerted when it breaks

A small LLM verifies each piece of evidence against your description. Failure → email, Slack, or webhook within minutes. Matrix view shows the cadence.

What gets monitored

Per-task evidence verification

Each task has an expected_evidence description. The agent submits text evidence; a small LLM verifies it matches what you described. Optional: a local verify_script runs through the MCP and submits its stdout instead.

Cadence detection

Workflow scheduled daily/weekly/cron? We notice when a run fails to start by its deadline + grace window. No more 'wait, has that agent run this month?'

Drift alerts in real time

Email, Slack incoming-webhook, or generic webhook. Configure per workflow per alert type. Hear about a failed verification within minutes, not weeks.

SOC 2-grade audit trail

Every run, every submitted evidence, every verifier verdict — preserved per your retention tier. CSV + JSON export. Reviewable trail when compliance asks.

Workflow matrix view

Tasks across columns, runs down rows, status icons per cell. Five seconds from open to 'I see something is wrong.' Drill into any cell to see the actual evidence.

Embeddable status badge

Drop the SVG badge in your README. Every README view = a passing/failing pulse on your agent's health, visible to your team and the world.

Your agent claimed "task complete".

Was it? agent-check is the layer above per-call observability — workflow process monitoring, MCP-native, with AI-validated evidence per task.

Make sure your AI agents actually finish the job.