Orca is an agent collaboration platform that sits above existing AI coding harnesses like Claude Code, Gemini CLI, and Codex, turning individual agents into a coordinated development team. It provides persistent agent identities, scoped file ownership, structured communication channels, and centralized credential management so developers can mix and match runtimes per agent. The platform is currently in beta and was used to build itself with zero human-written production code.
Funding
Funding not disclosed
Founders
Product
Problem
Developers using AI coding agents like Claude Code, Gemini CLI, or Codex typically work with single, isolated agents that lack persistent identity, clear ownership boundaries, and structured communication. This limits collaboration between agents, creates credential management risks, and prevents teams from scaling AI-assisted development beyond individual tasks.
Solution
Orca is an agent collaboration layer that sits above existing harnesses, turning individual AI agents into a coordinated team. It gives each agent a persistent identity with a name, role, and instructions, and scopes ownership to specific files and directories so no single agent manages the entire codebase. Agents communicate through structured channels, asking questions, delegating work, and staying aligned on shared goals. The platform supports multiple runtimes simultaneously, allowing developers to mix Claude Code, Gemini CLI, Codex, and Copilot CLI per agent. Credentials are centrally managed and injected at runtime, keeping keys out of repositories. Orca also includes work management features for assigning tickets, tracking progress, and managing sprints across the agent team.
Target Audience
Primary users are software developers and engineering teams who already use AI coding harnesses and want to scale collaboration across multiple agents for larger, more complex development projects.
Features
- Persistent agent identities with configurable names, roles, and system instructions
- Scoped file and directory ownership to prevent single-agent codebase control
- Structured agent-to-agent communication channels for delegation and alignment
- Multi-runtime support enabling simultaneous use of Claude Code, Gemini CLI, Codex, and Copilot CLI
- Centralized credential vault with runtime injection and zero-access proxying on the security roadmap
- Ticket assignment, progress tracking, and sprint management for agent teams