Your projects.Your context.Your agents.

Claude Code, Codex, Cursor or any other agent that speaks MCP work from the same decisions, instructions and project knowledge. Each one gets exactly the part that concerns it.

Request early access See how it works

Works with

Claude Code
Codex
Cursor
Copilot
Grok
Ollama

The problem

Every agent starts from nothing.

Claude Code
Makes a call on the API.
Codex
Knows nothing about it.
Cursor
Asks you all over again.
Files
CLAUDE.md, AGENTS.md, .cursor/rules: one per agent.

esfis decides which context each agent gets, and keeps a record of where it came from.

What it looks like

One decision in Claude. Codex already knows.

Claude picks PostgreSQL for the API. esfis stores it with a date and a source, and the next day Codex has it.

1Claude decides

Claude Codeopus

› Postgres or Supabase for the API?
● weighing the options…
  PostgreSQL: it is already in production
  and the migrations are yours.
◆ esfis · save the decision

2esfis stores it

esfisdecision

The API runs on PostgreSQL. Supabase ruled out.

space
project1

source
Claude Code

date
14 Sep 2026

status
current

3Codex receives it

Codexnext day

› build the users model
◆ context · decision
  The API runs on PostgreSQL
  Claude Code · 14 Sep
✓ db/migrations/001_users.sql

How it works

Exactly the context that belongs there.

With a source, an expiry and permissions. Nothing from another project.

1
Agents write to it
When Claude makes a decision, or you give it an instruction, it saves it to esfis over MCP. Nobody has to open a file.
2
esfis keeps it in order
Each item carries a type, a space, a source, a date and an expiry. When it stops being true, it stops being delivered.
3
Each agent gets its own
Codex works in project1, so it gets the context of project1 and your personal context. Nothing from project2 ever reaches it, however much that project also talks about databases.
4
You can see why
Every delivery is recorded. esfis --why tells you what went in, what stayed out and the reason.

Claude Codeopus

› Postgres or Supabase for the API?
  PostgreSQL: it is already in production.
◆ esfis · save the decision

decisionThe API runs on PostgreSQL

esfis✓ saved to project1

esfisdecision

The API runs on PostgreSQL. Supabase ruled out.

spaceproject1
sourceClaude Code
date14 Sep 2026
howstated outright
validuntil superseded

One piece of context, with where it came from and how long it holds.

Codex asks for context in project1

decisionThe API runs on PostgreSQLproject1
instructionEndpoints follow /api/v1/resourceproject1
preferenceAnswer in Catalanpersonal
technologyProject2 runs on MongoDBproject2

Terminal

$ esfis --why 212
delivery 212 · codex · project1
✓ in   The API runs on PostgreSQL
✓ in   Endpoints follow /api/v1/resource
✓ in   Answer in Catalan
✕ out  Project2 runs on MongoDB
        another space

Context, not memory

More than memory. Context.

esfis does not try to remember everything: it gives each agent what it needs to work on this project.

A preference of yours holds everywhere. A client's decision never leaves that client's space.

Already have a CLAUDE.md or an AGENTS.md? They still work: esfis can write them for you with --compile.

Memory

“I prefer TypeScript.”

True for you, anywhere.

Project context

technologyNext.js and PostgreSQL
decisionSupabase ruled out: the migrations are ours
instructionEndpoints follow /api/v1/resource

True inside project1. Decided by Claude Code yesterday.

The app

A macOS app to run it all.

Projects, agents, terminals and tasks, all around the same context.

Project1

Claude Code opus

› add the pricing page
◆ context · decision
  Prices are shown without VAT
● reading app/checkout/…
✓ app/pricing/page.tsx
✓ components/Plan.tsx

Tasks

Review the checkout
Error-path tests
Update the README

Codex gpt-5.5

› review the checkout tests
◆ context · technology
  Next.js and Postgres
● 3 files changed since yesterday
✓ 42 tests passing

claude 42% codex 18% CPU23% RAM61%

What the app gives you

Projects
A folder you already have becomes a project, with a context space of its own.
Agents
Real terminals, several at once, with your PATH and your keys.
Tasks
A queue beside the agent. It lives in .esfis/feina.md, inside the project.
Activity
Files, Git, sessions and the latest decision in every project.

Local by default

Your context is yours.

It is stored in ~/.esfis, on your disk. The MCP server opens no ports.

You decide which agent sees which space, and afterwards you can check what each one received. No account needed.

This Macno open ports

~/.esfis/
├─ memory.db   context and spaces
└─ MEMORY.md   readable copy

Who can see the project2 space

Claude Codehas access
Codexno
Cursorno

No agent should ever start from nothing again.

The esfis app is still in development.

Request early access