Formally verifiable, observable agent that safely self-improves on diverse sets of tasks. Software that writes itself safely and reliably.
Current Status
Currently nearly 100 github stars, not yet marketing, pre-release/polish product phase. Currently taking VC meetings with groups like Pebblebed and super{set} to raise pre-seed funding to build out core infrastructure.
Problem or Opportunity
Currently agent harnesses are static - they do not remember user preferences or improve on tasks over time. This results in user frustration, token cost inefficiency, and sub-optimal outcomes on performed tasks.
Solution (product or service)
Ploke allows the agent harness to autonomously improve itself to better fit the user's profile and work domain with custom tools, skills, and prompts that update themselves as the user's demands shape the agent's self-improvement.
Business model
GTM strategy flexible and needs evaluation, but several options:
- revenue by serving models through application (gcloud + openrouter already wired) and taking cut of token overhead.
- enterprise offerings: create custom agents using Ploke's core self-improvement system for specific use-cases
- (experimental) use blockchain-based technology to reward users that train agents on diverse tasks, collect subscription from end-user consumers
- (experimental) collect formally verifiable specifications and agent traces to sell to big labs
