Rakazo: Open-Source Grok Bot Alternative with Voice Mode and Persistent AI Teammates

To move beyond the structural limitations of disposable AI chat sessions where environmental context, credentials, and configuration state reset after every prompt, open-source engineers are turning toward persistent runtime architectures. Developed by Elie (GitHub: elie222) and highlighted across the engineering community by Kanika (@KanikaBK), Rakazo is an Apache-2.0 licensed, full-stack platform built as an open-source alternative to Grok Bot, designed to operate autonomous, long-running AI teammates on self-controlled infrastructure.

Image source: GitHub elie222/rakazo / @KanikaBK on X

Deployable locally with a single Docker command or scalable across isolated cloud environments such as E2B and Daytona, Rakazo preserves authentication sessions, conversation histories, and routine Markdown specifications, providing a dependable collaborative foundation for engineering workflows.

Full-Stack Architecture on Self-Controlled Sandbox Infrastructure

Rakazo avoids proprietary vendor lock-in by executing entirely on infrastructure governed by the user or organization. Developers can host the platform on bare-metal local machines, containerized Docker instances, or ephemeral virtualized sandboxes like E2B and Daytona.

• Modular Service Layout: The backend is divided into a high-performance web API built on Hono (apps/api) and a dedicated background job runner powered by Graphile Worker (apps/worker).
• Data Persistence and Execution Engine: Data models and system states are maintained through Prisma ORM backed by a PostgreSQL database, while core agent tasks run through the PiAgentRuntime execution environment.
• Single-Command Docker Deployment: The entire infrastructure stack—including the API server, background workers, and persistent database—can be launched locally using a single Docker execution command.

To protect the host operating system from untrusted or unpredictable code generated by foundation models, all execution paths remain strictly confined within Docker containers or E2B sandbox boundaries.

Persistent Workspaces, Multi-Agent Delegation, and Cross-Platform Clients

Unlike standard code-generation assistants that discard working state once an execution loop concludes, Rakazo treats environment persistence as a foundational requirement for production collaboration.

• State-Preserving Sandboxes: Each agent instance retains its browser login states, historical conversational logs, and task definitions stored in Markdown files. Work can be paused and resumed without re-authenticating services or rebuilding directory context.
• Broad Operating Privileges: Agents are equipped to navigate headless browsers, issue terminal commands, manipulate local filesystem structures, and interact with desktop GUIs, enabling end-to-end automation spanning codebases and external web portals.
• Peer Delegation Architecture: Rather than funneling all logic into a monolithic agent prompt, Rakazo implements horizontal peer delegation. Agents can dynamically instantiate short-lived subagents assigned to specialized subtasks, coordinating parallel progress before aggregating output.
• Multi-Platform Access: Teammates can interact with and supervise running agents across a unified workspace available as a web client, an Electron desktop application, and an Expo mobile application.

Real-Time Voice Interfaces, Tool Ecosystem, and Deployment Considerations

Rakazo extends beyond keyboard-driven chat by integrating real-time voice workflows that mirror spoken engineering standups and pairing sessions.

• Voice Pipeline Integration: By integrating API keys from OpenAI, ElevenLabs, and Cartesia, the platform supports real-time two-way voice calling, spoken dictation, and low-latency synthesized vocal feedback.
• Extensible Integration Protocols: System capabilities can be extended through Composio and Pipedream Connect, alongside native compatibility with remote Model Context Protocol (MCP) servers and standard OpenAPI definitions.

Teams planning deployments should take note of several verified engineering constraints. Rakazo is currently in an active beta stage, meaning API interfaces and configuration schemas may undergo breaking iterations. System administrators must ensure Docker or E2B isolation remains strictly active to guard against unrestricted host access. Additionally, all underlying model inference and speech synthesis resources require Bring-Your-Own-Key (BYO) configuration or self-hosted model backends.

Sources

The complete source code, architectural specifications, and release discussions for Rakazo are documented across the primary repository and initial community announcement:

• Rakazo GitHub Repository (elie222/rakazo)
• Kanika X Announcement Post (@KanikaBK)