Yup, I definitely faced multiple loops or "blocked" scenarios when iterating on the prompt and tools.
The most obvious one is when the Agent generates a terminal command that requires some interactivity (ex: "confirm to ...").
This causes the Agent to build all files from scratch adding a 5x more steps than necessary.
I'm planning to try to make terminal STDIO work as a "human in the loop" pattern or even better as an "Agent in the loop" one.
Claude Code is a complete tool that is ready to be used locally, focusing on Anthropic API features, while this coding agent is for educational purposes (like OpenAI's Swarm) and works with any LLM provider.
Thought this Coding Agent could also be used in a real setting as a Pull Request or GitHub Coding Agent!
1. Yes, we currently provide a simple approach to retries but are already working on providing features for idempotency.
2. All sensitive data (tokens, env vars) are encrypted on our side, however, we don't prevent users to print their values in the logs yet - it's planned in our next releases.
I see. I hope you have a great journey. I used to want to implement a system like that for my import/export tasks because my workload heavily involves importing and exporting CSV files. But the deadline defeated me ;) I have no time to design a system like that
By the way, I really love how simple the defer API is to use
We aim to provide a similar feature set as Sidekiq (throttling, unique jobs) but with a complete hosted solution.
While Defer and Temporal can both be used for writing background jobs, workflows, and CRONs, there are some core design differences.
Temporal has been created as the Kubernetes of highly distributed systems, enabling developers to write code that runs on multiple regions without worrying about possible termination of the program and interruption of workflows spanning across multiple steps.
While Temporal can be used for background jobs, workflows, and CRONs, its main goal is to ensure that highly distributed tasks will reliably be executed.
That's the main reason why Temporal API is so verbose, with many concepts to deal with.
Defer, on the other hand, provides comparable reliability while focusing on the developer experience.
You can write workflows, CRONs, and jobs that run for hours without worrying about them being terminated.
All this, with a simple API that enables you to write some workflows (background functions calling other background functions) in plain TypeScript, with no mental model to fit in.
When deployed in non-local environments, background functions get executed on our infrastructure with their (securely stored) arguments, spawn in a dedicated isolated temporary container.
For our users concerned about data locality, we recommend pushing the minimal data as arguments (ex: ids or external ids) and fetching the data during the execution on our side, if needed, through a dedicated SSH tunneling setup.
Once an execution is done, its associated isolated container - gets a dedicated VPC and disk - gets destroyed permanently.
We will also provide on-premises solutions for Enterprises.
In local environment (dev), background functions run completely synchronously and locally, no call is made to Defer.
I'm planning to try to make terminal STDIO work as a "human in the loop" pattern or even better as an "Agent in the loop" one.