The user's Anthropic Claude account was permanently suspended for violating the Usage Policy without specific details. An appeal resulted in a rejection, and the official data export provided only the user's name, email, and phone number, excluding all chat history and project data.
The author recovered data from a manual backup created on August 12 and CLI session transcripts found in WSL. To continue using the Claude Code client, the user configured a LiteLLM proxy to route requests to the Grok model, bypassing the client's model name verification.
To avoid future vendor lock-in, the user migrated their workflow to a self-hosted infrastructure using Open WebUI, Postgres, and the Ollama engine. The system now supports multiple providers, with data stored locally and tools integrated via MCP.
The new setup allows for independent model switching and centralized management of chat history and project files. The user also migrated existing ChatGPT data and implemented a custom tool for managing context window limits and token compression.
The author plans to implement a unified routing system to handle provider failovers and limit management. The current stack relies entirely on open-source components, with only the AI models remaining as external dependencies.
