And it's alive!! Excited to share a new OSS framework I've been cooking that gives you the building blocks to build your own agent harness:
- vendor agnostic hooks
- automatic session isolation
- automatic change provenance via Jujitsu
- cross repo task tracking
https://t.co/9hDLYhpwhc
Ships with a prescriptive workflow I've been dogfooding for the last six weeks, but is built to be completely customizable to fit how you want to work.
Let me know what you think!
Pre-built binaries and https://t.co/vD4BjeYJVO coming soon...
@connoratlunon@ycombinator Building the “Terraform of agentic engineering” that provides persistent context, guaranteed execution, and universal controls across all agent harnesses. Modeled from learnings leading product at HashiCorp and Tailscale
@realmcore_ Persistent memory of task state across sessions can make retrying, etc possible. Has been a pain to get reliable orchestration because of harness idiosyncrasies but I’m close
@bhalligan This is one of the reasons I built a “tldr” command for Aiki. Not reading every diff is okay, not understanding how things work is a recipe for disaster.
@diptanu The walled gardens are real, but I don’t know if they have another choice. Funding has pushed them to be everything to everyone.
Luckily enough of us are old enough to remember AOL.
@ivanburazin I’m betting the house on it. Actively building the inner loop by shifting automation into the coding workflow and tracking provenance of changes as they happen. GitHub will be reduced to a system of record, once the system of work arrives.
My general thesis is that multi-cloud is just beginning for AI, but unlike the last era it will be an operational requirement instead of a risk mitigation.
Companies will need a neutral infra layer that enables shared context, execution control, and governance so that their teams can experiment with different models/harnesses at will.
Aiki is that control layer