Product notes, engineering decisions, and field lessons from the Helix team.
Close the session and the AI forgets who you are? Dream Mode makes Helix review finished conversations, distill your preferences and project knowledge, and actually use them in the next session.
Once AI starts reading and writing your codebase, "where did the data go" stops being an abstract question. Here's how Helix uses three connection modes to keep Agents running on your own infrastructure.
HelixVM puts AI agents inside a lightweight local VM sandbox — secure isolation without buying cloud resources, installing VMware, or learning virtualization.
Helix treats worktree isolation as a system-level constraint. Agents never touch main directly — every change lands on an isolated branch, passes a code review gate, and only then merges back as a single clean commit.
Manager Mode is Helix's multi-agent orchestration architecture for complex engineering work — a Manager, an Execution Agent, and parallel SubAgents with strict separation of duties, scope locking, and evidence-based completion.
Helix treats the model as the current engine of a session, not the identity of it. Users can switch models anywhere in a conversation — default model, mid-session, or even per tool call — with the full history preserved, no restart required.
Helix isn't yet another AI that can write code. It's a product that treats AI agents as a real engineering system — multi-agent orchestration, parallel execution, durable context, and cross-workspace delivery.