cognira

What is cognira?

cognira is a self-learning AI chat assistant built by solo developer Thomas Conway. You teach it a preference, fact, or word, and it saves that to private memory instantly. Every reply shows which memory it used, with a one-tap Forget button. It runs on its own model, served from a single iMac.

Top Features:

• Visible memory: each reply lists which saved memories it used for that answer.
• Guarded learning loop: shared knowledge updates only after several independent users teach it.
• Local Entity app: run the same models on your own computer with local memory.

Use Cases:

• Personal preferences: teach it your tone and formatting rules once for future replies.
• Private drafting: brainstorm names, outlines, and short answers without ads or data sales.
• App building: call the OpenAI-compatible API to add the model to products.

Who Can Use cognira?

• Privacy-minded users: people who want an assistant that shows exactly what it remembers.
• Developers: builders who want a low-cost OpenAI-compatible endpoint for their apps.
• Writers: anyone who wants short, consistent replies that follow taught style rules.

Pricing

• Free (£0): a weekly allowance of 25K tokens plus the local Entity app.
• Pro (£15 per month): raises the weekly allowance to 500K tokens for daily use.
• Max (£30 per month): gives 2M tokens every week for heavy use of the assistant.

Pros and Cons

Pros:

• Transparent memory: you always see which saved fact shaped each answer you get.
• Independent model: it never forwards your questions to an outside AI provider.
• Same model everywhere: paid tiers only add usage allowance, never a different model.

Cons:

• Compact model: frontier assistants still beat it on broad knowledge and trivia.
• Solo project: one developer builds and runs it, so support may be limited.
• Data stored: conversations are kept so eligible parts can feed its learning.

FAQs:

1) Does it really learn from my chats?

Yes, taught facts and preferences apply from the very next reply.

2) Is it a wrapper around another model?

No, it runs its own model built on a Gemma base.

3) Can I run it offline on my computer?

Yes, the Entity app runs the same models locally on your computer.

4) How much does the API cost?

It costs 1.50 pounds per million tokens with a five pound minimum.

5) Does it show ads or sell data?

No, there are no ads and user data is never sold.