Open source first.
The desktop app, releases, issues, and GPLv3 license all live on GitHub. Developers can inspect how SharkBay launches agents, tracks status, stores task records, and syncs context before putting it in their workflow.
Launch Claude Code, Codex, Gemini, Kiro, Qwen, OpenCode, and other coding agents from one local project workspace. SharkBay is GPLv3 open source, leaves your agent instructions intact, keeps teamwork context in Git, and turns finished artifacts into shareable snapshots when you choose to publish them.
Launch and watch supported coding agents from the same project workspace.
The desktop app, releases, issues, and GPLv3 license all live on GitHub. Developers can inspect how SharkBay launches agents, tracks status, stores task records, and syncs context before putting it in their workflow.
SharkBay works alongside project instruction files such as AGENTS.md and CLAUDE.md instead of editing them. It keeps status hooks limited to visibility: they report what the agent is doing so SharkBay can show working, approval, and stopped states without replacing your existing agent habits.
Teamwork context is stored through Git, including the sharkbay-team-context branch, and authentication follows the developer's existing GitHub and gh command setup. The goal is low coupling: your repositories, identity, and collaboration history remain portable outside SharkBay.
Add the repositories you care about, launch agents in project context, run dev services, review task records, generate artifacts, share finished snapshots, sync teamwork context, and keep browser tabs attached to the same workspace.
Register exact local directories and switch between them from the sidebar.
Open supported coding agents in visible project terminal tabs.
Track agent state, terminal activity, dev commands, browser tabs, and Git changes.
Use task records so another agent or teammate can understand what changed and how it was checked.
SharkBay Task Protocol writes concise Markdown records under each project. Agents can read the same scope, files, work notes, and verification history without depending on a private chat transcript.
The first advantage is not "we support many CLIs" in the abstract. SharkBay turns those CLIs into visible project tools: pick a repo, start the right agent, and keep the session attached to that workspace.
Multi-agent work breaks down when every terminal looks equally alive. SharkBay makes agent state a first-class signal, so users know where attention is needed before opening a tab.
The island is the "ambient awareness" feature. It is not another dashboard; it keeps long-running agent work visible while the developer is in a browser, editor, or terminal elsewhere.
SharkBay's collaboration story is deliberately Git based. Context is not trapped in one agent's chat window or a SharkBay account; it becomes task records and team memory that move through the repository's context branch.
The protocol gives agent work a durable shape: scope, files, work notes, verification, and commits. It is lightweight enough for local development, but structured enough for handoff.
## Summary Implemented the homepage mockup for review. ## Files - docs/mockup/index.html ## Work - Built a static SEO-friendly page structure. - Linked product actions back to GitHub. ## Verification - Opened the local mockup in a browser. - Checked responsive layout and links.
Review is a separate mode, not just another agent prompt. SharkBay starts a constrained session that reads task context and writes a local report instead of making changes.
When the task is visual or web-facing, the browser belongs next to the terminal. SharkBay makes local preview and agent output part of the same project surface.
Some agent work is meant to be opened, inspected, and handed off: mockups, reports, docs, and single-page tools. SharkBay gives agents a scoped Artifact output path so the finished result stays tied to the task that produced it.
When an Artifact is ready to leave the local machine, SharkBay can publish it as an unlisted share link. The snapshot shows the output without exposing the repository, task records, or local workspace.
SharkBay treats dev servers as project state, not random background commands. The user sees which service is running and can get back to its logs quickly.
Agent work should never feel detached from Git state. SharkBay puts branch, dirty files, and recent activity in view before the user commits to a direction.
Sometimes the right move is a small edit beside the terminal. SharkBay gives users enough file access to inspect and patch context without becoming a full IDE.
Long-running agent work often spans interruptions. SharkBay keeps sessions visible enough that users can identify, restore, and continue the right thread.
The website should make the trust model obvious: SharkBay is open-source desktop software, not a hidden cloud workspace. It coordinates local agents while keeping project instructions and login outside SharkBay's ownership.
Agent sessions keep running after you step away from the Mac. SharkBay can pair with your own self-hosted Telegram bot, so you can check in, steer agents, and clear approvals from your phone.
Long agent output and busy web previews are easier to navigate when you can search them in place.
The public site keeps the product story simple. Downloads, source code, release notes, issues, and GPLv3 license details stay on GitHub where developers already work.