shadok-ai
shadok-ai is an open-source web cockpit for running real Claude Code sessions side by side, each in an isolated git worktree, using the user's existing Claude subscription. It adds browser and optional Telegram control, scheduling, profiles, secrets management, quota pacing, and review-oriented workflows.
At a Glance
- Individual developers using Claude Code
- Software engineering teams coordinating multiple coding agents
- DevOps, monitoring, and repository-maintenance workflows
- Teams needing Telegram-based remote agent supervision
- +1 more
AI Tools by shadok-ai
(1)
shadok-ai
Claude Code Sessions Manager
Discussions
No discussions yet
Be the first to start a discussion about shadok-ai
Latest News
Promote to 0.10: twenty-three merges reach the beta channel; named accounts, invitations, prompt authorship, session-token/capability recovery, self-reload continuation, ledger persistence, phone tour, and favicon improvements.
The lead reads the project it is actually in: project-ground detection now identifies repository conventions, version control, stack markers, CI, and convention files deterministically.
Alpha package reached 0.10.1 following the project-ground change.
v0.7.0 first tagged release: scheduled agents and a no-model guard, plus Telegram, child-agent reporting, profiles, isolation/review, and live-preview improvements.
Products & Services
Install with npx shadok-ai. It launches a local browser cockpit and drives multiple real Claude Code TUI sessions in parallel, with one process/channel per agent, isolated git worktrees, interactive dialogs, diffs, profiles, secrets, scheduling, quota gauges, self-update, and optional Telegram control.
Recurring or one-shot prompts can run a deterministic shell guard before the model. No stdout means the agent is not started and zero model tokens are spent; output wakes the agent with the result, while guard errors raise an alarm.
An optional Telegram bot maps each agent to a topic, turns permission and multiple-choice dialogs into inline buttons, supports prompts and attachments, and exposes commands for spawning, scheduling, secrets, restart, and update.
Market Position
shadok-ai positions itself as an open-source, local-first orchestration cockpit around real Claude Code TUI sessions, rather than a hosted agent platform. Its distinctive angle is combining parallel worktree-isolated sessions and human-in-the-loop controls with deterministic guards that avoid model calls when checks are quiet; the release notes contrast this subscription-driven approach with the documented programmatic lane of the Claude Agent SDK/API-key model.
Leadership
Founders
Alexandre Cognard (GitHub: gnarco)
The original npm package metadata identifies gnarco as maintainer and the package's early publisher; the shadok-ai repository's main contributor is the shadok-ai-dev account. Public profile evidence identifies gnarco as Alexandre Cognard, Arianee CTO & Co-Founder in Paris; Arianee describes him as an expert in blockchain and distributed systems who leads its protocol R&D and enterprise-platform technical architecture.
Executive Team
Alexandre Cognard
Creator / maintainer (GitHub: gnarco)
Public project/package attribution connects gnarco to the project; his public profile identifies him as Alexandre Cognard. He is also identified by Arianee as CTO and an expert in blockchain and distributed systems, leading protocol R&D and enterprise-platform architecture.
Founding Story
The project was started as a self-hosted cockpit for coordinating multiple real Claude Code processes rather than repeatedly driving one agent. Its central idea is that scheduled checks should run without waking the model when there is nothing to report, while agents can work in parallel, communicate completion or questions, and leave reviewed changes. The project says much of its own code and specifications were built by agents running inside the cockpit.
Business Model
Revenue Model
Open-source, self-hosted software distributed through npm under AGPL-3.0-only. The cockpit runs on the user's machine and uses the Claude CLI and the user's existing Claude subscription; the project documentation says no API key is required and nothing leaves the machine by default.
Target Markets
- Individual developers using Claude Code
- Software engineering teams coordinating multiple coding agents
- DevOps, monitoring, and repository-maintenance workflows
- Teams needing Telegram-based remote agent supervision
- Self-hosting and Docker users
- Running several Claude Code tasks concurrently on one codebase without worktree collisions
- Scheduled repository monitoring and reporting where quiet checks should cost zero model tokens
- Remote approval and supervision of coding agents from Telegram
- Human review of agent-generated changes before delivery
- Giving agents bounded roles, command guardrails, and selected credentials
- Self-hosted development automation on a laptop, server, or Docker container