Open-source tools for coding agents · Ankara, Türkiye
I build tools for keeping project memory, connecting agents, and checking what their work actually proves. Start with one small task and inspect the result.
Website · X @okgrit · Instagram @okgrit · Instagram @ismt.aydin · Get in touch
Your agents. Your work. One workspace.
My main project is Everhew, a local-first workspace for AI agents. I'm designing it around keeping tasks, conversations, approvals and results connected across desktop, mobile and terminal.
Start here: Try done-is-a-claim on your next small task.
Follow this profile for coding-agent tools and concrete usage notes. You can also find me at @okgrit.
Your agent says “done”. What would prove it?
Incident-backed rules and focused skills for coding agents, with local tools for capturing and rechecking evidence. Start with a Markdown rules file; add the workflows you need.
Use it on one task · Read the rules · Try the demo
I already told you.
Shared local memory for AI agents, with source-linked records and correction history. Available as a source alpha; native-client integrations are still being validated.
Try a first save and recall · Client evidence and limits
A small mailbox between your ChatGPT dot and Claude Code.
Send a request, claim the work, and read the reply through ordinary HTTPS. Python standard library, with no model API calls. The coding session does the work; you decide what it may authorize.
How it works · Trust and security
Parcel to Design · P2D
Tools for turning parcel and zoning inputs into architectural plans, sheets, renders, and editable CAD files. Architect review required.
Useful tools. Inspectable work.
I also build websites and AI automations. Have a concrete problem to build around? Let's talk. Türkçe de yazabilirsiniz.
Case note 1: parsing a report must not erase its exit code
A synthetic Bash reproduction: a report can fail while its JSON formatter succeeds. See the observed results and the pipeline fix.