Agent Field Report: Private Multi-LLM Agents — Week of 2026-09-03
This week, our air-gapped portfolio rebalancing agent, powered by private multi-LLM agents running entirely on local Ollama, executed a critical rebalance of our SOL/USDT position. The result: a 3.2% gain on the rebalanced portion, bringing our portfolio back into target allocation, all while ensuring zero market-sensitive data ever left our local network. This deployment highlights a growing differentiator for serious traders: the ability to automate sophisticated strategies without exposing positions or logic to third-party databases.
The Setup
We deployed a specialized portfolio rebalancing agent designed to maintain a strict 60/40 target allocation between Solana (SOL) and USDT. The agent was built using Assistant Hub's Strategy Lab and configured to run on a dedicated local server, isolated from the public internet. Our chosen LLM for this task was Llama 3 8B, hosted via Ollama, ensuring all computational and decision-making processes remained entirely on-premises. This setup was chosen specifically to demonstrate a fully air-gapped trading operation, where privacy of trade signals and portfolio composition is paramount.
The agent's mission was straightforward: continuously monitor the SOL/USDT pair and, should SOL's percentage within the portfolio deviate by more than 5% from its 60% target, trigger a rebalance. For example, if SOL's allocation dropped to 55% or surged to 65%, the agent would initiate a buy or sell operation to restore the 60% weighting. Our initial portfolio stood at