Private beta memodis is in private beta - onboarding a small group of pilot teams. Apply for access →
Tacit knowledge → explicit graph

Capture the knowledge your team never wrote down - for humans and agents.

Every decision in Slack, every architecture call in a meeting, every "we tried that, it didn't work" - memodis extracts them into a structured knowledge graph that humans search and AI agents query. No new wiki to maintain. No more re-discovering.

For teamsCapture · retrieve · share
For agentsMCP-native context API
Built in EUGDPR-first · audit-logged
◆ The cost of knowledge loss

When the person who knew leaves, the project goes with them.

Studies put it at 20–40% of institutional knowledge lost per year to turnover, async sprawl, and silent doc decay. Most of it never made it into a wiki because nobody had time. memodis closes that gap automatically.

◆ Tacit knowledge → explicit graph

Watch unstructured work become a queryable graph.

Scroll. memodis extracts decisions, people, and facts from the surfaces where work happens, and writes them to a graph in real time.

Slack · #eng-infra · Tuesday 14:22
JM
Jamie Mendez14:22 quick q - why did we pick Postgres over the obvious Dynamo? Was that documented anywhere?
AK
Alex Kovac14:24 Cost. Dynamo's read-pricing model at our query volume would've been ~$8k/mo. We modelled it. Pinning to Postgres + read replicas hit $1.2k. Decision was August '24, I ran the spreadsheet.
RT
Rita Tan14:25 Also: our hot-path queries are joins. We'd have ended up doing app-side joins in Dynamo, which is the worst of both worlds.
Meeting · Eng All-Hands · Mar 6
AK
Alex Kovac08:14 "…and the second thing - we're never touching Kafka for this. We tried it on the last project and the operational burden ate two engineers for a quarter. For our scale, Redis streams is enough."
SM
Sara Moss08:16 "I want this documented. Next person to suggest Kafka should have to read why we chose not to."
GitHub · PR #1042 · review comment
RT
Rita Tancomment Reverting this - we already tried using the connection pool for read replicas in v2. It deadlocks under reconnect storms. Use the per-replica pool pattern from db/replica.ts instead. Postmortem in the ops doc from October.
AK
Alex Kovac+ context ↑ This is the kind of context that lives in two people's heads and nobody else's. memodis catches it.
Notion · Engineering / ADRs · Feb 2026

Why we ship feature flags before features.

Author: Sara Moss · 4 min read · 12 references

Every customer-facing change ships behind a flag, off by default, even if we plan to enable it the same hour. The cost is a few lines of code and a Statsig record. The upside is a fast undo if something breaks under real load.

Why this matters

Three of our last four production incidents were "rollback took 45 minutes because we shipped a database migration and an API change in the same PR." A flag would have made the rollback a one-click toggle.

This isn't a process tax. It's a rationality discount on future incidents.

Support · Customer call · Acme Corp · Mar 14
AC
Acme · Customer22:14 "We've been seeing really bad latency on the dashboard, but only on weekends. Friday evening through Monday morning. The same query that's 80ms on Tuesday is 4-5 seconds on Saturday. Is this a known thing?"
RT
Rita Tan22:18 "That's our analytics cron colliding with the read-replica pool. We saw it once in November. Fix is to move the cron to a dedicated replica - we have the design doc, just hadn't shipped it. I'll prioritise it this sprint."
Six weeks later · new engineer onboarding
Why aren't we using Kafka? ⌘K
memodis · graph answer

The team decided not to adopt Kafka in March 2026. Two reasons documented in the graph:

1. Operational burden: on the previous project, Kafka ate two engineers for a quarter on operations alone.

2. Scale fit: at current load, Redis streams meets the need. Authority: Alex Kovac (then-tech-lead), confirmed by Sara Moss in the same meeting.

meeting · all-hands · mar 6 person · sara moss person · alex kovac fact · 2 eng / quarter
Sources processed: 5
Decisions 0 People 0 Facts 0 Alex Kovac Decision: Postgres $8k vs $1.2k / mo Sara Moss Decision: not Kafka Burden: 2 eng / quarter Rita Tan Conn-pool deadlock Decision: flag-first Principle: ship fast Acme · customer Weekend latency 4-5s Edges: 0
◆ Who memodis is for

Two kinds of memory loss - one solution.

For companies

When the person who knew leaves and takes the project with them.

Series A through 500-person orgs lose 20–40% of their institutional knowledge to turnover and async sprawl. memodis catches it before it walks out the door.

  • Onboarding without 3 weeks of pairing
  • Decisions you can audit 2 years later
  • One canonical answer instead of three wikis
  • Cross-team retrieval, ACL-respected
For AI agents

When your agent has the model - but none of your team's context.

An agent without your memory is a smart intern on day one. memodis plugs into MCP, Claude, Cursor, LangGraph - so the agent reads the same room your team does.

  • MCP tools for retrieval, lookup, trace
  • Stable entity IDs across multi-step runs
  • Audit log of every agent query
  • Self-hosted option for regulated stacks
◆ How it works

Three steps from messy to retrievable.

Step 01

Connect your sources

OAuth into Slack, Notion, Drive, GitHub, Linear, Zoom - or use the ingest API. memodis indexes incrementally; originals stay where they are.

Step 02

memodis builds the graph

Decisions, entities, projects, people - extracted with cited sources. Review and correct anything that's wrong. The graph keeps itself in sync.

Step 03

Ask - humans or agents

Search via web, Slack, CLI, or MCP. Every answer ships with the trace back to the source. Citations, not hallucinations.

◆ Security & data

Built for teams that care where their data lives.

Hosting

EU only (Frankfurt)

Encryption

AES-256 + TLS 1.3

Audit log

Every query, every actor

Compliance

GDPR · SOC 2 in progress

Private beta · pilot program

Apply to be one of our pilot teams.

We onboard a handful of companies each week. Tell us a bit about your team - we reply within 48 hours.

You're on the list.

We'll be in touch within 48 hours from hello@memodis.com. Watch the spam folder just in case.