The problem
An agent’s useful work includes its conversation, files, and decisions. If those live only with a running worker, stopping or changing the execution environment makes that work difficult to recover. Exploring a second approach also needs a way to carry the context forward without mixing the results together.
Rat Things is an Indubitably engineering project for operators who want the execution service and saved work in their own AWS account.
The approach
Separate durable session history and workspace storage from the worker performing a task. Saved conversations and results can remain available after that worker is gone. The operator controls the deployment, including its accounts and storage configuration.
A prepared session can fork into independent directions. Each fork starts with the shared context and workspace, then develops its own work. A native Rust console gives the operator a place to follow those sessions alongside the API and command-line tools.
A working example
The demo below follows five design directions from one shared brief. It shows real session outputs, their comparison, and the removal of the four unchosen sessions. The video is an edited 42-second walkthrough; its duration is not a measure of execution speed.
What the project demonstrates
- Conversation and workspace retention can be separated from a worker’s lifetime.
- Independent session forks let an operator compare approaches that share the same starting context.
- API, command-line, and native console access can work around the same saved sessions.
Boundaries and lessons
Saved work does not mean every running process survives interruption. Execution recovery depends on the environment, and retention still needs a separate backup policy. Permissions are resolved before a run; operations outside that boundary remain unavailable.
This is an engineering preview for owner-operated deployments. Broad browser reliability, sustained-load guarantees, and use with untrusted tenants remain outside the project’s claims. The case study documents implemented behavior rather than client revenue or time-saving results.
Explore the evidence
Read the project overview, the capabilities and boundaries, and the source repository.