Monitoring and Audit

Every action your agents take is visible and auditable. This isn’t optional, it’s built into how Moe’s works.

Real-Time Monitoring

Telegram / Slack

Your primary monitoring interface is the messaging platform your agents use. Every agent-to-agent message appears in a topic or channel:

#research — research briefs and findings
#qa — verification results and screenshots
#ops — status updates, deployments, coordination

You can watch agents work in real time from your phone. No dashboard required.

What You See

For every agent action:

Who — which agent performed the action
What — the specific action taken (search, edit, commit, screenshot)
When — timestamp
Why — the brief or instruction that triggered it
Result — what was found, changed, or verified

Intervention

Redirect

Reply to any agent message to change its course:

Research Agent: "Found 3 sources. Proceeding with synthesis."
You: "Also check their LinkedIn page for recent announcements"
Research Agent: "Good call. Checking LinkedIn now..."

Pause

Tell an agent to stop what it’s doing:

You: "@content hold on, don't commit yet"
Content Agent: "Holding. Ready when you are."

Override

Correct an agent’s output before it moves to the next step:

Research Agent: "The price should be $500 per trip"
You: "Actually it's $450 for groups of 4+. Update the brief."
Research Agent: "Updated. Passing corrected info to Content Agent."

Audit Trail

Built-In History

Telegram and Slack both provide searchable message history. Every agent decision, every handoff, every human intervention is recorded with timestamps.

Structured Logs

Each agent also writes to its own memory/YYYY-MM-DD.md file with details that don’t need to go into the group chat:

API calls made
Sources consulted
Confidence levels
Error recovery actions

Compliance

For regulated industries, the combination of messaging history and agent memory files provides:

Who decided what — clear attribution to specific agents
Human oversight proof — every intervention is timestamped
Data lineage — which sources informed which decisions
Change history — git commits for all content changes

Alerts

Configure agents to escalate specific situations:

Confidence below threshold — “I’m only 60% sure about this, flagging for review”
Error after retries — “API failed 3 times, need human help”
Scope creep — “This task is bigger than expected, should I proceed?”
Sensitive content — “This involves financial data, requiring approval”

Alerts can go to Telegram, Slack, email, or SMS depending on urgency.