DeepSeek published a design for an agent with no sessions. We built a small one from that design, gave it a real repository with a real bug, and recorded what it did. This page replays that run.
What is real: every model decision, tool call, pull request, CI result and crash recovery below. What is staged: the bug is planted, the team's lines are scripted and sent by a presenter, the "laptop" is a separate cloud container, "next morning" is simulated by clearing the Router's window, and the Before screen is a simulation. This is a small demonstration, not a product.
You open a session for each task, paste in what it needs, and remember what each one knows. In DeepSeek's experience, a person can follow only 3 to 5 active sessions at once.
Nothing connects these sessions except you. The session about the pull request has never heard of the login bug, even when they are the same problem. A simulated screen, but you have probably lived it.
Press Play. On the left is what the team sees. On the right is what the agent did, as it logged it at the time.
A Router: one long-lived session on DeepSeek V4.1 Flash that talks to people and dispatches work, and has no shell. Workers: one Modal container per task on DeepSeek V4 Pro, with its own clone of the repository. A filter: code rules first, then a model reading the team's policies. Threads with subscriptions, so follow-ups skip the Router. Inline & Spill and Chain of Recall, so old detail stays behind IDs. An asynchronous loop, a session database, a small laptop daemon and a Watchdog.
It follows the design in DeepSeek Harness's note, which is theirs. Our code and any mistakes in it are ours.
Built and written by Dr. Raj Dandekar and the team at Vizuara AI Labs, for a webinar on 4 November 2026. The design it follows is explained in full at superagent.vizuara.ai.
We teach with love, care and passion. If you want to build agents like this end to end, our live cohort Agentic Systems Engineering takes you from the LLM to retrieval, tools over MCP, the agent harness, workflows and evals, on a real business system. It is open to engineers moving into agentic AI roles and to enterprise teams.