Your code. Your bot.
One accountable AI workspace.
Bring selected files to DSHREM, compare two model answers, and control your workspace from your own Telegram bot. Every paid task has a hard limit and a visible receipt.
Only the files you choose
Attach code and screenshot text in VS Code. Private RAG retrieves your uploaded evidence with file paths and line references.
One model or a two-model council
Choose an available model, use deterministic routing, or request DeepSeek + GLM review. Both council calls share one task ceiling.
Spend, holds and credit
Know what was charged and what remains reserved. Unknown charges remain held. A failed paid call is never retried automatically.
Our methods, with clear boundaries
Evidence-first
Tenant-isolated lexical RAG, bounded source excerpts, hashes, cited paths and exact-request answer caching.
Plan → check → answer
Structured reasoning prompts and an optional second-model critique. These are workflow controls, not proof of superior model quality.
ROYAGENTS & integrations
LangGraph client integration and bounded council answers are available. Autonomous server changes, full worker fleets and automatic provider-project provisioning remain managed pilots.