Our take: Best for users wanting transparent assistant memory, privacy-minded users wanting a local model, and small-model enthusiasts who prefer independent, non-API-forwarded services.
Main limitation: Free plan limited to 25K tokens per week; pricing shown in GBP only (not USD); served from single iMac limiting scalability
By the AI Tool Atlas editors.
Based on the vendor's official documentation; we have not tested this product hands-on. What “documented overview” means
Who should use cognira
Best for users wanting transparent assistant memory, privacy-minded users wanting a local model, and small-model enthusiasts who prefer independent, non-API-forwarded services. Based on the vendor's documentation, not on a hands-on test.
Best for: Users wanting transparent assistant memory and visibility into what the AI learned
Also a good fit for
Privacy-conscious users wanting a local model option
Consider another tool if need high token allowance on free tier
Or if
Require USD pricing (GBP only)
Need guaranteed high-availability/cluster deployment
cognira pricing and usage limits
Prices checked Sep 29, 2026 on the official US pricing page. Official pricing ↗ Prices and limits change often.
Vendor pricing page re-checked automatically 12 days ago: no changes detected.
The official cognira pricing page as we saw it. cognira.dev, captured Oct 1, 2026
Free access and trial conditions
Free plan is £0 forever with no credit card required. Includes 25K tokens per week across both input and output. Same full model and features as paid tiers; no degradation or lite version.
Free plan serves as indefinite trial; all features available including local cognira Entity and learning
Pricing shown in GBP only; cancel anytime, keep access until renewal
For: Heavy users, obsessive thinkers, frequent chat and Proto users
Limits
2M tokens per week (4× Pro, 80× Free allowance)
Includes both input and output tokens
Includes
Everything in Pro, plus:
4× weekly token allowance (2M vs 500K)
3 more
Same full model quality
Heavy daily use support
cognira Entity included
No API
Usage caps, credits and extra costs
API available separately: £1.50 per 1M tokens (OpenAI-compatible endpoint); £5 minimum top-up required
Not mentioned; pricing shown in GBP
What is cognira, and how does it work?
cognira is an independently served chat assistant running its own model, cognira Retrain 12B, from a single iMac. Users can teach it preferences and vocabulary; each reply shows which memories were used, with a one-tap option to forget them. Teaching works immediately through a guarded learning loop that saves teaching immediately while weight updates wait for checks. cognira Entity is a local binary for Windows, macOS and Linux that runs the same models offline. Proto turns an idea into an editable branching map using the same weekly token allowance. The service is built and maintained by a single independent developer.
What you provide and what you get back
What you give cognira and what comes back, from its documentation
cognira Retrain 12B (proprietary, retrained on Gemma 4 base via MLX); Omega 188M deprecated
Hosted: single iMac; Local: Windows/macOS/Linux via cognira Entity binary
Private (per-user immediate), Shared (cross-user validated, word meanings only), Model weight updates (for future versions)
Weekly reset (Mondays); counts both input (user) and output (model) tokens
OpenAI-compatible; £1.50 per 1M tokens
GBP only (not USD)
Where cognira is most useful
Scenarios are worked out by our editors from the documentation; the actual result depends on your input and plan.
Everyday chat with persistent preferences
Who
Daily users, casual chatbot users
You provide
Chat messages with preferences and corrections
Expected result
Replies that reflect taught preferences; memory attribution shows which memories influenced the reply
Plan needed
Free (25K tokens/week) or Pro (500K tokens/week) depending on usage
Limitation
Free tier limited to 25K tokens/week; teaching limited to personal memory (shared knowledge requires multiple users) source (opens in a new tab)
Teaching custom vocabulary and facts
Who
Users wanting to train AI assistant
You provide
Corrections, preferences, new words/meanings, domain-specific knowledge
Expected result
Private memory stored immediately; eligible items become shared knowledge after validation by multiple users
Plan needed
Any tier (teaching is immediate on all plans)
Limitation
Only word meanings can become shared knowledge (not personal facts or preferences); shared knowledge requires corroboration from 3+ independent accounts source (opens in a new tab)
Mapping an idea into branches with Proto
Who
Planners, researchers, project managers
You provide
Idea outline with branches and sub-points
Expected result
Editable branching map using same weekly token allowance as chat
Private memory with per-reply memory attribution and Forget
Saves user-taught preferences, corrections, facts, and vocabulary to private memory immediately; each reply shows which memories were used; one-tap Forget button to remove any memory Users see exactly what the AI learned from them and can manage their own memories transparently
Access: Included on all tiersLimit: Private memory only; shared knowledge requires validation from multiple independent userssource (opens in a new tab)
Learning dashboard with pass/fail states
Shows public log of all teaching attempts (passed and failed) at cognira.dev/learning; displays per-user teaching leaderboard showing who has taught the model most Transparency into what the model learned system-wide and rejected
Access: Included on all tiers; teaching activity tracked publiclyLimit: Transparent to other users; teaching activity contributes to public leaderboardsource (opens in a new tab)
cognira Entity local app for Windows, macOS and Linux
Sealed local binary running same models as hosted chat (Retrain 12B and Omega); inference, memory, and optional weight-level retraining all stay on disk; nothing leaves the device Privacy and offline capability; users can run the full AI system locally without internet
Access: Included on all plans (Free, Pro, Max)Limit: Weight-level retraining optional; offline use limited by local inference speedsource (opens in a new tab)
Proto idea maps
Convert ideas into editable branching maps using the same weekly token allowance as chat; maps show structure and relationships Organize thinking and planning at the same cost as chat
Access: Requires account (all plans)Limit: Uses shared token budget with chat; not a separate feature tiersource (opens in a new tab)
cognira: strengths and limits
Reasons to choose cognira
Free plan with no card required Low barrier to entry; 25K tokens/week allows casual usedocumentedsource (opens in a new tab)
Same model on every plan No 'lite' model on free tier; answer quality is identical across Free, Pro, and Max tiersdocumentedsource (opens in a new tab)
Local desktop version available on all plans Privacy and offline capability included even on free tier; no extra cost for cognira Entitydocumentedsource (opens in a new tab)
Shows which memories influenced each reply Transparent AI behavior; users understand why the assistant responded a certain waydocumentedsource (opens in a new tab)
Check before paying for cognira
Free plan limited to 25K tokens per week Casual use only; serious daily use requires paid plandocumentedsource (opens in a new tab)
Prices listed only in GBP Non-UK users must convert currency; no USD pricing discloseddocumentedsource (opens in a new tab)
Served from a single machine Limited availability; vendor explicitly avoids cluster deployment for transparency and cost reasonsdocumentedsource (opens in a new tab)
Account needed for Proto maps and teaching Requires signup; no anonymous usedocumentedsource (opens in a new tab)
How to use cognira for using cognira for everyday chat with persistent memory
Set up cognira
Visit cognira.dev and sign in with email (Clerk handles auth)
Start a new chat and begin typing
Teach the model by correcting replies, rating them, or explaining new words
Memory is saved immediately and used in the next reply
Optionally: download cognira Entity for local offline use (same models, stays on disk)
Example: using cognira for everyday chat with persistent memory
Chat: 'Any name ideas for my hiking newsletter?'
cognira: 'Trail Notes — short, and easy to remember'
View which memories influenced the reply (e.g., 'Keep answers short', 'Call me Tom')
Teach it: 'Actually, I prefer longer, descriptive names'
Next reply will use this new preference immediately
One-tap Forget button available for any memory next to each reply
Review and export the result
Each reply shows 'Used 2 memories' or similar, with memory names listed
Memories appear in Settings → Learning
Teaching activity visible in public learning log at cognira.dev/learning
Local Entity: verify models downloaded and inference runs offline
Example prompt suggested by our editors (not a recorded output of the product)
Keep answers short. Call me Tom. I need names for a hiking newsletter.
cognira integrations and API
Plan comparison on the official pricing page. cognira.dev, captured Oct 1, 2026
Native integrations and export formats
Export: Chat transcripts (downloadable), Memory exports (implied via Settings), Proto maps (editable branching format)
API access and separate billing
An API is available, billing not documented. OpenAI-compatible endpoint; named API keys, usage graphs, flexible top-ups API documentation ↗
Privacy, data use and commercial rights
Summarized from the vendor's current policies; this is not legal advice and a vendor's badge is not an independent audit.
How your inputs and outputs are handled
Conversations stored for learning eligibility; private memories saved to account; teaching activity tracked. Analytics includes full chat content sent to PostHog. Teaching activity is public on leaderboard.
Eligible chat turns can become training candidates for future model versions; word meanings can become shared knowledge after validation from 3+ independent accounts; personal facts/preferences never eligible for sharing
Conversations stored indefinitely for learning; can be deleted on request via email
Commercial use and plan restrictions
Permitted; no restrictions stated
No privacy differences between tiers; all data subject to same privacy policy
Controls for business users
No ads, no data sales (stated)
Private teaching works immediately; shared knowledge requires validation
Teaching activity publicly logged but anonymized for promotion
Analytics opt-out available (off by default for product improvement)
Chat content sent to PostHog for analytics; input values masked in replays
cognira Entity is the local product — a self-learning AI that lives on your machine. Same models as hosted chat (Retrain 12B and Omega), but inference, memory, and optional weight-level retraining all stay on your disk. Nothing leaves. Included on every plan. source (opens in a new tab)
Does cognira really learn from my chats?
Yes. Teach it a preference, a fact, or a word and it's saved to your private memory immediately — the next reply that uses it tells you exactly which memory it drew on, with a one-tap Forget button. Some taught words can become shared knowledge after validation by 3+ independent users, but only after being reworded so it's never anyone's literal sentence. source (opens in a new tab)
How smart is it, honestly?
It is a compact, locally served model, and frontier systems will still beat it on breadth and trivia. Its edge is the loop: confirmed teaching works immediately from private memory, then verified updates can improve the model where people actually use it. source (opens in a new tab)
Is cognira a wrapper around someone else's model?
No. cognira does not forward your question to an external API. cognira Retrain 12B is our model, retrained on top of Gemma 4 base and served ourselves through MLX. cognira Entity runs those same models on your computer. The earlier Omega 188M checkpoint is open-source on Hugging Face. source (opens in a new tab)
How is it free?
The compact local serving setup keeps costs low. The Free plan covers casual weekly use; Pro and Max raise your weekly allowance and help pay for serving and the guarded learning loop. Every tier gets the same answers. No ads, no selling your data. source (opens in a new tab)
Not verified: USD pricing equivalent (GBP only: Free £0, Pro £15/mo, Max £30/mo); US availability or regional restrictions; SLA or uptime guarantees (single-machine limitation); Team/multi-seat options.
How we documented and maintain this page
Every fact about cognira above comes from the 3 official pages listed here, last compared on Sep 29, 2026; we have not run cognira ourselves, so it has no editorial score. How we check and re-check pages.
Article updated Oct 10, 2026; prices checked Sep 29, 2026. Outbound links to vendors are plain links; we do not earn commissions. How we review · Report an error
