I launched Anneal (annealit.ai) in alpha yesterday: an intelligence layer that sits invisibly between you and any LLM (OpenAI, Anthropic, Mistral, local models, etc.).

It gives your AI persistent structured memory, enforceable rules, teachable skills, and zero-latency context that compounds over time. Think of it as annealing metal: apply heat and patience so your AI gets stronger, more reliable, and truly yours. No vendor lock-in or opaque black-box memory.

Key features:

- Model-agnostic proxy; route calls once, switch models freely, keep all context
- Private by design; end-to-end encryption with your keys, air-gapped/enterprise ready (HIPAA, SOC2/PCI, defense use cases)
- Rules + Skills; define consistent behavior and reusable capabilities that persist across sessions and tools

Would love honest feedback on:

- Real-world use cases (personal assistant, enterprise workflows, agent building?)
- How it compares to the other memory layers you're using
- What rules/skills you'd actually want to teach your AI

Looking forward to the discussion and happy to iterate fast based on your input.

NPM package launching next week and B2B play for Google Workspace, Slack, Linear, and more coming shortly after.

Paper has been written, and full state spec will be AGPLv3'd at some point.

(Disclosure: This is my project / our team's work.)