invite only

Your sessions produce more
than you can hold.

Decisions, findings, dead ends, directions. Pragmus is where that goes — structured automatically, readable by your next agent.

Session knowledge has nowhere clean to go

Decisions scatter

Important calls end up buried in chat logs. You remember making the decision but not what you actually decided.

Dead ends repeat

Your agent proposes what you already rejected. Not because it forgot — it never knew.

Sessions produce more than notes can hold

The thinking that happened in a session — what was tried, what failed, what was resolved — disappears when the context closes.

Dump the session. Structure emerges.

Connect Pragmus via MCP. After each session, write what happened. Pragmus extracts, organizes, and builds structure automatically. The knowledge accumulates. You never maintain it.

Start each session fresh. Your agent reads structure, not history — not the full chat log from last time. Save tokens. Start clean.

01

Memory tools and knowledge graphs aren’t built for this

Vector stores recall fragments. Graph databases require traversal queries. Neither gives your agent a navigable structure it can orient in immediately.

02

Not a knowledge graph. A hierarchy.

Pragmus is not a graph database. It’s a self-organizing hierarchy — project, area, topic, pragma. Your agent drills down. No graph queries. No edge traversal. Just structure that already makes sense on arrival.

03

Structure emerges automatically

Every item is classified by intent — decision, feedback, directive, idea. Items connect to each other. Contradictions are detected. The result is organized knowledge, not a memory dump.

04

Organization is automatic

No folders. No tags. No maintenance. Knowledge self-organizes into areas and topics. Summaries update as items accumulate. The structure evolves with your project — you don’t manage it.

05

Your agent reads the graph

Via MCP, your agent navigates areas, topics, and summaries — not a flat list of memories. It understands what was decided, what conflicts, and what’s still open. That’s the difference.

What built up. Always there.

Your agent navigates, not searches.

When your agent connects, it reads pre-organized structure — areas, topics, summaries. It doesn’t search. It navigates what’s already there.

Reads are database queries. No LLM calls. Nothing on top of whatever you’re already running.

Zero additional LLM cost on reads
mcp · new session
query "where did we land on the pricing model?"
Three tiers — free, standard, pay-as-you-go.
Free tier has structural caps: 6 areas, 4 topics.
Standard is fixed monthly.

Open question — pay-as-you-go
floor relative to standard not finalized.
► 1 open question · 7 items · 0 LLM calls

People who think with AI

founders

Strategy that builds

Every positioning discussion, competitive insight, and product decision goes into Pragmus. Months of sessions become organized knowledge your next agent can draw from.

builders

Projects that accumulate

What was tried, what failed, what was rejected — organized once, there always. Your agent knows what came before.

agents

Sessions that compound

Running agents on a project? Each session adds to the same organized record. No re-explaining. No starting over.

Your sessions, organized.
Your work, preserved.

Works with any MCP-compatible client

Claude Code
OpenClaw
Cursor
Windsurf
VS Code
Zed
n8n
Zapier

No schema. No config. Free to start.

MCP-native · self-organizing · zero read cost

join the waitlist

Currently in closed beta. Sign up to reserve your spot.