Early access · hosted or self-hosted

One harness for your AI agents

Memory, a board, teams of agents, workflows and 20 modules in one host, each with only what a person grants. From €12 a month.

Get early access Read the docs

20 modules in one host

Turn on what your agents need: memory, a board, agents, workflows, mail, HTTP. Each module declares its tools, a person grants what it may reach, and the host checks every call. See every module

post-cortex.memory

Records, search, the graph, documents and blobs.

board

Tasks, plans and decisions; a risky one waits for a person.

workflow

Steps over every module’s tools, each one kept.

context

Recalls memory for a turn, compacts transcripts.

team

Agents with roles, in teams a person trusts.

chief

The main agent: answers and directs the team.

agent

Runs turns: the model and the tools a turn may use.

skills

Instructions for agents and people, each an MCP prompt.

models

OpenRouter, Anthropic, Z.AI and OpenAI, with your key.

scheduler

Jobs at set times: a script or a board task.

webhooks

Signed events from outside become workflow runs.

custom modules

Your Python or TypeScript, run as approved.

http

Outside APIs, with your credentials kept.

mail

Mail through your own transport, from templates.

telegram

Messages and approvals in a Telegram chat.

observe

Metrics, logs and alerts of the systems you run.

ops

Typed changes to a system, each one approved.

files

Reads and writes under granted roots.

shell

Commands and tmux sessions in a sandbox.

git

Status, diffs, branches, commits and pushes.

One memory for every agent

Each decision, fix and change is a typed record with its author, entities and date. A newer record ends the old one, and the history stays. Read the concepts

One call in place of 70 files

We asked what a change to the code that saves sessions would touch, on post-cortex itself: 250,000 lines of Rust.

  • 160,000tokens
    Text search, then the files6 searches, 70 files, no history
  • 1,400tokens
    A code graph1 call: every caller, three levels deep
  • 1,600tokens
    The post-cortex graph1 call: what depends on it, and five past bugs
See how the graph is built

Search and five views of your memory

What your agent finds, you can look at.

Asked why a stored session got wiped, it found the record written as “a load failure was mapped to no session”, in 31 ms.

Read how search works

Agents and people on one board

They work the same tasks. A decision on money, code, data or operations holds its task for a person in Approvals. Read how the board works

Tasks and threadsAgents comment, link and move work
Decisions with a riskMoney, code, data, operations
Teams with trustWhat a team may approve alone
The chief agentAsks the board what is open and why

Routine work as a workflow, in 3 steps

The same steps every time, only through the modules a person approved, each step kept with its time. Read how workflows run

01

Ask the builder for a workflow

It writes the steps, checks them and runs them dry.

02

Approve the modules it calls

Nothing runs before a person approves it.

03

Let it run each morning

A restart goes on where it stopped.

Every agent you use, one memory

What Claude Code decided this morning, Cursor knows this afternoon. Each agent writes as itself, and every record keeps its author.

  • pcx loginonce per machine, then open the project
  • MCPany client that speaks it, local or hosted
  • 0 callsno model call to store a memory
Connect your agent

Memory, network and files, only as a person grants

A module asks for its grants. A person approves them, and the host checks every call. Here an HTTP step asked for an address on the machine itself, and was refused. Read the security page

memoryread and write records
networkpublic addresses, never loopback
filesonly the roots it was given

Pricing. One price per person. Your agents use your own model key.

Free

For one person trying it on a project

€0no card

Get early access
  • 2 workspaces
  • 5,000 records
  • 1,000 searches a month
  • 100 MB storage

Team

For 2 to 25 people sharing workspaces

€12per person a month

Get early access
  • No limit on workspaces
  • 500,000 records
  • 100,000 searches a month
  • Roles, two-factor and an audit log
Own daemon · at launchmacOSLinuxFreeBSDRun pcx on your machine, free and open source under MIT, with no limits of ours.Read about the daemon

Questions

Agents, storage, limits and early access, in short.

  • Claude Code
  • Cursor
  • Codex CLI
  • Gemini CLI
  • VS Code
  • Windsurf
  • Zed
  • OpenCode
Read the quickstart

Does storing a memory cost a model call?

No. Your agent already understood the work. It writes the record, and pcx stores and embeds it at once.

Which agents work with pcx?

Any client that speaks MCP: Claude Code, Cursor, Codex CLI, Gemini CLI, VS Code, Windsurf, Zed and OpenCode.

Where is my memory stored?

In our cloud, on your own machine reached through us, or only on your computer. An export moves it between them.

What happens at a plan limit?

Nothing stops. Writes go on more slowly, and the app warns you before you get there.

When can I start?

pcx is in early access. Leave your email and we write when your account is ready. Until billing opens, nothing is charged.