Open source · Apache 2.029

Coding agents that remember how you work.

Prokop is an open-source agentic coding workspace. Persistent agents keep their own memory, skills, and searchable history across projects, while each project keeps its context separate and inspectable.

$ curl -fsSL https://prokopai.dev/install.sh | bash
installed to ~/.prokopai/bin/prokop
installed to %USERPROFILE%\.prokopai\bin\prokop.exe
$ prokop init
daemon started · client at http://localhost:8742

sessions, projects, files, terminals, permissions · same workspace on desktop and phone

The Prokop workspace on desktop: sessions, agent chat with tool activity, and file diffs
The Prokop client installed as a PWA on a phone

the problem

Your coding agent is smart.
Its working environment is still temporary.

Six sessions become six windows.

Parallel work turns into terminal management. The desktop becomes the job.

Every project starts from zero.

Context lives inside one session. Close it and the decisions, fixes, and conventions evaporate.

The new session has never met you.

Re-explain your stack, your conventions, your preferences. Every time. To a capable stranger.

Remote means a chat bridge.

Check in from your phone and you get a stripped-down bot, not your actual workspace.

continuity

Stop briefing a stranger every session.

Persistent agents keep their memory, skills, and history across projects. Project knowledge stays with the project. Turn on automatic learning to preserve useful lessons from your conversations. Knowledge stays in plain files you can read and edit.

agent memory · follows the agent
~/.prokopai/agents/reviewer/MEMORY.md
# how I like reviews
- short summary first, details after
- flag real issues, never rewrite style
- always link the failing test

Works in every project. It knows you, not just this repo.

workspace memory · stays with the project
~/krakatit/.prokopai/MEMORY.md
# project facts
- bun, never npm
- auth uses dayjs, not Date
- tests: smallest target first

Separate on purpose. Your repo's quirks don't leak into other work.

search: "flaky auth test"
session · fix-auth-flow · 2d ago
session · token-drift · 11d ago
session · ci-green-again · 24d ago
scope: this workspace, or everything an agent has done

Full-text search over past sessions. Recover decisions instead of redoing them.

SKILL.md · release-checklist
name: release-checklist
description: when cutting a release
1. run client + server checks
2. update VERSION files
3. changelog entry, then build

Skills preserve repeatable ways of working. Write them, or let the agent propose them.

automatic learning
turn it on. keep working.
review conversations, preserve lessons
optionally create and refine skills
no scheduled job to set up

Adjust the model, focus, or timing when needed. Inspect sources and knowledge diffs, or undo a change.

not model training · markdown files on disk · read, version, edit, delete

the workspace

Run the work, not the windows.

Sessions, projects, files, Git changes, terminals, and permissions in one focused interface. Any number of sessions running; the board keeps six live panes in view.

The Prokop workspace: session board with parallel agent sessions, files, and diffs

Board

Parallel sessions side by side. Six live panes, no window management.

Workspaces

Multiple projects with fast switching, plus an Overview that aggregates sessions across them.

Files and Git

Edit files, manage worktrees, create and switch branches. Browse commit history and diffs. Commit, pull (fast-forward), push, and rebase from the interface.

Terminals

Persistent terminals for long tasks. Keep talking while commands run.

Permissions

Every tool call visible. Approve one action, the session, or the workspace.

Schedules

Run recurring agent tasks with your own prompt and timing.

the scale

As plain as a CLI. As far as you want.

Everything is opt-in, per workspace. One project and a plain agent is a perfectly good setup. Turn capabilities on only if you want them.

01 / bare

One project. A plain agent.

Read, edit, grep, run. No memory, no skills, no search. Nothing runs that you didn't turn on. Start working immediately.

  • single workspace
  • built-in coding tools
  • nothing else on
workspace settings · capabilities
memory
skills
session search
learning
scheduling
everything else

02 / wired

The agent starts to know you.

Turn on automatic learning to preserve useful lessons from conversations. Memory survives sessions, skills capture repeatable work, and session search brings past decisions back into reach.

  • memory on
  • skills on
  • session search on
  • automatic learning on
workspace settings · capabilities
memory
skills
session search
learning
scheduling
everything else

03 / full send

A whole operation on one machine.

Session panel, six live sessions on the board, files alongside. Multiple workspaces under one Overview. Persistent agents with their own homes and memories, any number of sessions running, long tasks in background terminals while you keep working.

  • board: six live panes
  • multi-workspace overview
  • persistent agents
  • background terminals
six live sessions · click to view full size

on your phone

Not a Telegram bot. Your actual workspace.

The client is a mobile-optimized PWA served by Prokop itself. Same sessions, same agents, same history, same permissions. Mobile reorganizes the interface into one surface at a time. It does not reduce the product to a chat bridge.

  • The same agents and their memory, from anywhere
  • Approve permission requests from your phone
  • Push notifications when long runs finish
  • Reach it privately through networking you choose, like Tailscale

Prokop ships the PWA, auth, and optional TLS. The network is yours.

The Prokop workspace as a mobile PWA: sessions, files, and agent activity

in control

Familiar does not mean opaque.

Every tool execution is visible. Diffs land in the interface. Permission requests show the action, the risk, the resource, and the intent before you grant anything.

permission request
action shell · bun run test:server
risk medium · executes commands
target ~/projects/api
intent run the server test suite
approve oncethis session

Scopes you can revoke. Auto-approval boundaries you set per session or workspace.

changes · src/auth/session.ts
@@ session/token-drift fix @@
- const token = localStorage.getItem('auth')
+ const token = await store.getToken()
  if (!token) redirect('/login')
- setExpiry(Date.now() + 3600_000)
+ setExpiry(token.expiresAt)

Edits arrive as diffs. Files, memory, and prompts stay inspectable.

visible tool executionobserver + controller statesrevocable grantsauto-approval boundaries

yours

The working relationship stays yours.

The full stack is open source, Apache 2.0. Server and client. It runs as a daemon on your machine, independent of the browser or terminal. No account, no telemetry, no cloud gate between you and your workspace.

Open all the way through

Not open-core. The server, the client, the tools, the scheduler. Apache 2.0.

Files, not a database mystery

Memory, skills, and prompts are plain files on your disk. Version them, share them, delete them.

Bring your own provider

Connect supported providers or use Codex subscription authentication. Choose from the models available in Prokop.

Extensible by design

MCP tools over stdio or OAuth, browser automation, git worktrees. Tools appear alongside the built-ins.

model availability depends on the integration and prokop's supported configuration

questions

Straight answers.

Where does my data live?

On your machine. The server keeps its data and SQLite database under ~/.prokopai, and each project keeps its memory and skills as plain files inside the project directory. Nothing is uploaded to us. There is no us.

Does Prokop send telemetry?

No telemetry, no analytics, no account. Network traffic is what you set up: the model providers you connect, MCP tools you install, and the daemon serving the client on your machine.

Which model providers can I use?

Direct support includes OpenAI, OpenRouter, DeepSeek, MiniMax, Zhipu, and Zhipu Coding, plus Codex subscription authentication. Models from other companies can be reached through OpenRouter. Prokop does not add direct Anthropic or Google integrations of its own.

How is Prokop different from a CLI coding agent?

It is not a different agent. It is the workspace around one. Keep the model you already use and trade the terminal farm for a board of live sessions, files and diffs, persistent memory, session search, permissions, and the same interface on your phone.

What does Prokop cost?

Nothing. The full stack, server and client, is Apache 2.0. You pay your model provider for whatever you use. No seats, no tiers, no feature gates.

Can I actually work from my phone?

Yes. The client is a mobile-optimized PWA served by your daemon: the same sessions, agents, history, and permission approvals, reorganized for one surface at a time. You reach it over a network you control, such as Tailscale. Prokop ships the workspace, not the network.

Who builds and maintains Prokop?

One developer, Daniel Bilek. It is an active, evolving project rather than a hardened enterprise product, and the code is Apache 2.0, so it stays yours no matter what.

install

Bring your next project into Prokop.

One command, then prokop init. The daemon starts, the client opens, you pick a model and start working. Everything else is opt-in.

$ curl -fsSL https://prokopai.dev/install.sh | bash
installed to ~/.prokopai/bin/prokop
installed to %USERPROFILE%\.prokopai\bin\prokop.exe
$ prokop init
daemon started · client at http://localhost:8742