Skip to content

pre-release 0.1.4#66

Merged
xoxruns merged 13 commits intomainfrom
agent_iteration
Mar 30, 2026
Merged

pre-release 0.1.4#66
xoxruns merged 13 commits intomainfrom
agent_iteration

Conversation

@xoxruns
Copy link
Copy Markdown
Owner

@xoxruns xoxruns commented Mar 30, 2026

Multiple changes have been made to simplify the performance issues of certain architectural decisions and agentic iteration over the agent's memory.

  • switching to sqlite and removing PGvector. Not usable for fast multi-agent structures. Still maybe need to see if the installation is still done otherwise remove it. needs clean up
  • adding an agentic virtual filesystem with tooling to read, list, grep and write. grep is now used with ripgrep, needs testing.
  • The avfs can have different workspaces to work with. the idea today is that we have a memory workspace and the root one for future codebase analysis.
  • We've switched the simple python sandbox to a more usable component instead of a server like transport to the webassembly sandbox. We still need to check if the sandbox runs perfectly when built. We also should remove the download of the former implementation.
  • Adding a validation process with numerous exit strategies. Through pattern, llm-as-judge, tool backed or hybrid.
  • Adding report generation at the end of each successful vuln found.

xoxruns and others added 10 commits March 26, 2026 11:11
Moving the setup directory to environments.
First setup of an agentic Virtual filesystem, with read,write,ripgrep and list to gather information from memory.
I've also changing the realm where the agent operates. Now everything goes to .cache/deadend/agents/... and saves the agent id used in config so that we can gather the agent information directly.
the memory files are present inside that place. and they are simple MD files gathered after each agent run. So that the next agent knows what's going on.
resolving a small issue in read tool not converting correctly the lines into int for enumerate.
Removing relevance in most agent thoughts return. Those didn't serve any purpose.
registry was saving the models api keys as prints.
Implementing the tracing to agent thought.
    Some issues around the write on the avfs write tooling got stuck. It needed changes to a wrapper tooling to not let the model decide the workspace, which was a pretty idiotic thing to do. I mostly blame claude, but I need to move fast, so fuck this shit.
    Added changes to the instructions tools descriptions according to the tools added.
    the validation for now will be kept as is until after the platform launch.
really fucking hate coding agents. seems like things changed that I don't even know why. And what have been implemented is too fucked up.
@xoxruns xoxruns self-assigned this Mar 30, 2026
@xoxruns
Copy link
Copy Markdown
Owner Author

xoxruns commented Mar 30, 2026

Needs to close #58 #57 #32 #11 #36

@xoxruns xoxruns added bug Something isn't working enhancement New feature or request agent-iteration Agent iteration core changes labels Mar 30, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

agent-iteration Agent iteration core changes bug Something isn't working enhancement New feature or request

Projects

None yet

1 participant