AI CUSTOMER MEMORY
A Slack bot that turns customer-success notes into a persistent, queryable context layer instead of a trail of scattered docs, DMs, and stale handoff notes.
← Back to ProjectsThis anonymized portfolio version is based on a customer-success workflow problem: useful account context is created in Slack, but then disappears into scattered Google Docs, Notion pages, DMs, and personal memory.
The bot lets a CSM save notes, todos, recent updates, and account-history questions directly from Slack. Customer context is resolved from the channel name by default, persisted to a database, and queried with Claude when someone asks a question.
Portfolio focus: Workflow-native capture, shared customer memory, Slack product design, LLM Q&A over account history, and a demo path that mirrors a production deployment shape.
/csnote
Save account context from the customer channel where the work is already happening.
/csnote todo
Store action items alongside the customer history instead of letting them drift.
/csnote recent
Pull the latest notes before a call, handoff, renewal discussion, or escalation.
/csnote ask
Let Claude answer account questions from the stored notes, with dates and caveats.
Question for Northwind Logistics
What is the latest on their SSO rollout? I am jumping on a call with them shortly.
Answer from customer history
Initial setup was completed in March. They are on Okta using a hash-fragment RelayState format.
Pilot group went live with positive feedback, but a redirect-loop bug later affected the rollout.
Before the call: check whether ENG-1421 has shipped and follow up on the SAML attribute mapping doc.
Note for Northwind Logistics
Got off the call with them. They are happy about progress and want a couple of feature requests incorporated into the roadmap.
The point is not another note-taking surface. It is capturing context in the Slack channel where customer work is already happening.
Successful notes and answers post in-channel so the team shares the same context; errors stay private so typos do not clutter the channel.
The repo includes a Slack Bolt app, Socket Mode runtime, customer resolver, SQLite persistence, Claude-powered Q&A, seeded demo data, and tests.
The demo runs locally with SQLite, while the production shape maps to a long-running ECS/Fargate process with Postgres on RDS.
Channel-based customer resolution. A channel like northwind-channel becomes the default customer context, with an explicit override for cross-customer notes.
Persistent customer memory. Notes and todos are stored in a simple Postgres-shaped schema so the local SQLite demo can move cleanly to RDS.
LLM Q&A over full note history. Claude answers questions from stored notes and is instructed to cite dates, avoid speculation, and distinguish notes from todos.
Private runtime, public value. Socket Mode avoids a public inbound endpoint while still giving the team a Slack-native interface.