Agent Workflow Examples

These examples show how an AI agent can use Botmem's MCP tools to answer real questions by combining multiple tool calls.

Before applying connector or source filters, agents should call status or sources when they do not already know what data is available. For latest/current-state questions, use list sorted by eventTime instead of semantic search.

"What did John say about the project deadline?"

The agent needs to find a person and then search for relevant memories.

Step 1: Find the contact

Tool: search_contacts
Input: { "query": "John" }

Returns John Smith with contact ID contact-uuid-1.

Step 2: Search memories involving John about deadlines

Tool: search
Input: {
  "query": "project deadline",
  "contact_id": "contact-uuid-1",
  "limit": 10
}

Returns ranked memories from email, Slack, and WhatsApp where John discussed deadlines. The agent can synthesize a response:

"Based on your memory, John mentioned the project deadline in three places:

1. In a Gmail email on Jan 15, he said the deadline is March 1st (FACT, confidence 0.9)
2. In Slack on Jan 20, he mentioned pushing it to March 15th (UNVERIFIED, confidence 0.6)
3. In WhatsApp on Jan 22, he confirmed March 15th with the team (FACT, confidence 0.85)"

"Remember that the meeting was rescheduled to Friday"

MCP is read-only. To write a new memory, use the REST agent endpoint or CLI instead.

POST /api/agent/remember
{
  "text": "The team meeting originally scheduled for Wednesday has been rescheduled to Friday this week."
}

This creates a manual memory that will be embedded and enriched automatically. Future searches for "meeting schedule" or "Friday meeting" will surface this memory.

"Give me a timeline of the product launch"

The agent builds a timeline by searching and sorting.

Step 1: Broad search

Tool: search
Input: {
  "query": "product launch",
  "limit": 20
}

Step 2: For each result, check the event time and source

The agent receives memories from multiple sources and can construct a timeline:

Product Launch Timeline:

• Jan 5 (Email, gmail) -- Initial launch date discussion: "targeting March 1st"
• Jan 12 (Message, slack) -- Design review: "mockups approved, dev starting next week"
• Jan 20 (Email, gmail) -- Delay notification: "moving to March 15th due to API changes"
• Feb 1 (Message, whatsapp) -- Team sync: "beta testing starts Feb 10"
• Feb 10 (Photo, photos) -- Screenshot of beta dashboard tagged with launch project

"Who have I been talking to most about the budget?"

The agent combines statistics and contact lookups.

Step 1: Search for budget-related memories

Tool: search
Input: {
  "query": "budget financial spending",
  "limit": 50
}

Step 2: Analyze the entities in the results

The agent counts person entities across all returned memories:

"Based on your memory, the people you discuss budget topics with most are:

1. John Smith (23 mentions) -- mostly via email and Slack
2. Sarah Chen (15 mentions) -- primarily in email threads
3. Mike Johnson (8 mentions) -- WhatsApp group discussions"

"Is there any conflicting information about the contract value?"

The agent leverages the factuality system.

Step 1: Search for contract-related memories

Tool: search
Input: {
  "query": "contract value amount price",
  "limit": 20
}

Step 2: Check factuality labels

The agent examines the factuality field of each result:

"I found a potential conflict in your memory:

• Email from Sarah (Jan 15, FACT): 'The contract value is $250,000'
• Slack message from Mike (Jan 18, UNVERIFIED): 'I heard the contract is closer to $300,000'

The email from Sarah has higher trust (gmail: 0.95) and is labeled as FACT. Mike's Slack message is UNVERIFIED. You may want to confirm with Sarah."

"What was I doing last Tuesday?"

The agent browses the specific event-time range across all source types.

Step 1: Browse the date range

Tool: timeline
Input: {
  "from": "2026-02-25T00:00:00.000Z",
  "to": "2026-02-26T00:00:00.000Z",
  "limit": 20
}

Step 2: Supplement with location data if available

Tool: timeline
Input: {
  "from": "2026-02-25T00:00:00.000Z",
  "to": "2026-02-26T00:00:00.000Z",
  "source_type": "location",
  "limit": 10
}

Step 3: Check photos

Tool: timeline
Input: {
  "from": "2026-02-25T00:00:00.000Z",
  "to": "2026-02-26T00:00:00.000Z",
  "source_type": "photo",
  "limit": 10
}

The agent combines all results:

"Here's what I found for last Tuesday (Feb 25):

• 9:00 AM -- You were at the office (OwnTracks location)
• 10:30 AM -- Email thread about the product roadmap with Sarah
• 12:00 PM -- Photo at a restaurant (Immich, tagged: you + John)
• 2:00 PM -- Slack discussion in #engineering about the API migration
• 5:30 PM -- WhatsApp message to family group about dinner plans"

"Show me everything about my relationship with Acme Corp"

The agent performs a broad entity search and contact correlation.

Step 1: Search by organization name

Tool: search
Input: {
  "query": "Acme Corp",
  "limit": 30
}

Step 2: Find contacts associated with Acme

Tool: search_contacts
Input: {
  "query": "Acme"
}

Step 3: For key contacts, get their memory history

Tool: search
Input: {
  "query": "Acme Corp",
  "contact_id": "acme-contact-uuid",
  "limit": 30
}

The agent builds a comprehensive picture combining emails, meetings, messages, and contact metadata.