What Is Mansa

Mansa is the autonomous AI workforce Adullam Technologies built for BKR Capital. It runs on Anthropic Claude Managed Agents and handles the day-to-day work of processing inbound deal flow, so the BKR team can focus on decisions.

What Mansa Does

Mansa runs three scheduled workflows:

Daily Inbound Processing (7 AM Toronto time, weekdays)

BKR Capital receives roughly 10–20 emails per day to the shared inbox, with the majority being noise (newsletters, service provider pitches, irrelevant cold outreach). Every morning, Mansa:

1. Pulls the previous day’s emails from the BKR Capital shared inbox (pipeline@bkrcapital.ca)
2. Classifies each email as an investment lead, newsletter, spam, or general chatter
3. Deduplicates relevant leads against existing records in Attio CRM
4. Extracts structured data from email bodies, attachments, and pitch decks (including Google Slides)
5. Creates or updates company and contact records in Attio
6. Drafts a reply to the founder asking for a pitch deck, if one was not included
7. Sends each BKR team member a digest email summarizing what was processed

Reviewing and Sending Founder Replies

Mansa drafts replies but does not send them automatically. After the daily run, the BKR team:

1. Receives the digest email in their individual inboxes with a summary of what was processed
2. Opens each draft in Outlook, reviews it, and edits as needed
3. Decides whether to send the reply

This keeps a human in the loop for every outbound communication to a founder.

Post-Meeting Recap

After each Deal Intro meeting, Mansa:

1. Finds the meeting notes in Attio
2. Extracts every decision (Initial Appraisal, Watchlist, Low-Priority Watchlist, Pass) with the recorded reasoning
3. Updates each company’s stage on the Attio pipeline list
4. Assigns follow-up tasks to the appropriate team members in Attio
5. Emails the team a full recap with decisions and action items

Weekly Memory Compaction

Once a week, the Dreamer agent compacts each agent’s patterns.md memory store. It summarises incident logs into structured sections, collapses duplicates, and strips lead-specific data. This keeps agent memory focused and prevents performance from degrading over time.

How the Team Interacts with Mansa

Mansa runs entirely on schedule. The BKR team’s day-to-day touchpoints are:

• Reading the daily digest: arrives in each team member’s individual inbox each morning
• Reviewing reply drafts in Outlook: open the draft, edit if needed, decide whether to send
• Updating configuration: when the agent’s behaviour needs adjustment, update skills, rubric files, memory stores, or context files via the Claude Console

Key Concepts

Term	What It Means
Agent	An AI worker with a specific role. The four agents are Mansa (coordinator), Email Processor, Meeting Agent, and Dreamer
Skill	A reusable tool definition that agents can call (e.g. send-email, email-actions)
Rubric	A structured guide that tells the agent how to evaluate an inbound email or meeting decision
Memory store	A file store agents read from and write to across runs, recording what they’ve learned
Context file	A reference document (e.g. about_bkr_capital.md) agents read to write in BKR Capital’s voice
Session	One complete run of a workflow