Managing AI coding sessions across multiple projects has become a coordination problem that most developers solve with multiple terminal windows, separate browser tabs, and a growing sense of fragmentation. Agenttik, an open source project released under the MIT license, takes a different approach: a single workspace that runs Claude Code, Codex, and GitHub Copilot side by side, using the official CLIs and your existing login credentials.
The project is built with Go, SQLite, Vue 3, and Wails, and ships as both a desktop application for Linux, macOS ARM64, and Windows x64, and a web interface that runs on localhost. The core idea is that AI coding assistants should not each occupy their own silo. A developer working across several projects in a day should be able to see all their active sessions, switch between projects without losing context, and manage tasks from one place.
Multi-Project Sessions Without the Tab Soup
Agenttik organizes work by project. Each project gets its own workspace with its own session history, file tabs, and conversation state. You can archive and restore conversations per project, which means old sessions do not clutter the active workspace but remain available if you need to revisit a prior interaction.
The practical benefit is visibility. Instead of tracking which terminal window corresponds to which project and which AI assistant, you see all your projects in a single interface. Each project's sessions are independent, so switching contexts does not contaminate one project's conversation history with another's.
For developers who work on multiple repositories simultaneously, this eliminates the cognitive overhead of managing separate environments. The interface shows active sessions across projects, and you can jump between them without the friction of opening new windows or re-authenticating.
The Orchestrator for Cross-Project Work
One of Agenttik's more unusual features is the orchestrator, a separate pinned project that can inspect work and manage tasks across your other projects. The orchestrator has its own editable instructions and can be reset to a built-in prompt. It is designed for situations where you need to coordinate changes across multiple repositories or track a feature that spans several codebases.
The orchestrator is not a replacement for the AI assistants themselves. It operates at a higher level, looking at the state of work across projects rather than executing individual coding tasks. For teams working on microservices architectures or monorepos with multiple packages, this cross-project visibility fills a gap that single-project tools leave open.
Live Sessions and Prompt Queuing
Agenttik streams AI replies in real time, so you see output as it arrives rather than waiting for completion. You can resume previous conversations, which means a session that was interrupted by a meeting or a context switch picks up where it left off without losing the accumulated context.
The prompt queuing system lets you stack multiple prompts for a single project and let them execute sequentially. This is useful for chained tasks where one operation needs to complete before the next begins, such as running a test suite after a refactor or building documentation after code changes.
Model controls let you choose which model to use, set effort levels, configure permissions, and save favorite model-permission combinations. This avoids the repetitive process of selecting the same configuration every time you start a new session.
Code Tools Built Into the Workspace
Agenttik includes integrated code browsing and editing. You can open changed files alongside your project tasks, compare diffs, stage individual files, and prepare commits directly from the workspace panel. Markdown and HTML previews render inline, so you can review documentation changes without leaving the interface.
The Git integration is practical rather than comprehensive. You can view diffs, stage changes, and write commits, but the tool does not attempt to replace a full Git client. It covers the workflow you need when you are iterating on code with an AI assistant and want to commit the results without context-switching to a terminal.
Usage Tracking and Cost Visibility
Agenttik tracks token usage, context consumption, subscription allowances, and costs where the underlying tool reports them. For developers paying for API usage or managing team budgets, this provides visibility into how much each project consumes and which sessions are approaching limits.
The tracking is local and per-profile, so it does not require sending usage data to an external service. This aligns with the project's local-first philosophy, which keeps all state in a SQLite database on your machine.
Profiles and Private Mode
The profile system lets you maintain separate workspaces with isolated projects and tasks. A developer who works on both personal and professional projects can keep them completely separate, including conversation history and project metadata. Switching profiles happens through a picker before shortcuts, and each profile maintains its own state.
Private mode creates a temporary instance whose app data is removed on exit. Project files stay on disk, but the session history, settings, and database are discarded. This is useful for one-off tasks where you do not want the interaction recorded, or for testing Agenttik itself without contaminating your regular workspace.
Remote Connections and Server Mode
Agenttik can connect to a remote instance using a host:port address or HTTPS URL. The client checks the server's version endpoint before attempting connection and displays the server's login if authentication is required. This enables scenarios where the AI assistants run on a more powerful machine but you want to interact from a lighter client.
The server binds to localhost by default, which is a sensible security posture. The documentation warns that anyone who can reach an unauthenticated server has full control. Before exposing Agenttik through network settings, you should configure authentication and use a trusted network.
CLI Integration Philosophy
Agenttik does not wrap AI model APIs directly. It integrates with the official CLIs for Claude Code, Codex, and GitHub Copilot, using your existing login. This means you get the same model access, the same rate limits, and the same authentication you would have using the CLI directly. The difference is that Agenttik manages the sessions, provides the multi-project workspace, and adds the tooling layers on top.
For OpenCode Go, Agenttik supports direct use with a subscription key and an optional CLI. The settings default to the CLI when it is installed, falling back to the key-based approach otherwise.
Setup and Build
Running Agenttik requires Go 1.25 or later and Node.js 22.12 or later with npm. One of the supported AI CLIs must be installed and logged in on your PATH, or an OpenCode Go key must be configured in settings.
The quickest path is the web interface: run make run-web, open localhost in a browser, add a project folder, and start a session. The terminal command agenttik --init in a repository directory adds it as a project, whether or not the application is already running, and it appears in an open window immediately.
For the desktop app on Debian or Ubuntu, install native build dependencies with make deps, then run make run. Cross-compilation for Windows uses make build-windows-amd64. macOS ARM64 builds create an ad-hoc-signed application bundle and a ZIP archive. No Apple developer account is required, but downloaded builds are not notarized, so you will need to approve them through system settings after the first blocked launch.
What This Represents
Agenttik reflects a maturation in how developers interact with AI coding tools. The initial wave of AI assistants operated as isolated experiences: one model, one conversation, one terminal window. The next phase involves managing AI as a coordinated part of a multi-project workflow, with the same kind of workspace organization that developers expect from their IDEs.
The local-first architecture matters for teams with security constraints or data sovereignty requirements. All state lives in SQLite on your machine. There is no cloud sync, no telemetry, no external service dependency. The AI models themselves are external, but everything else stays local.
As AI coding assistants become standard tools rather than novelties, the infrastructure around them needs to support how developers actually work: across multiple projects, with multiple tools, in contexts where switching costs are real. Agenttik is one of the first projects to tackle that coordination problem directly.