Key Takeaways
- Free tier (Β£0, 25K tokens/week) plus paid Pro (Β£15/month, 500K tokens/week) and Max (Β£30/month, 2M tokens/week) β see Cognira Pricing and License below
- Proprietary, closed-source β no LICENSE file, public source code, or GitHub repository found
- Cognira Entity: a sealed local binary for Windows, macOS, and Linux, bundled with every pricing tier
- Hosted chat is served from a single 16 GB iMac via MLX, per cognira.dev β a small, single-machine operation rather than a data-center-scale service
- Built and run by one independent developer, Thomas Conway, not a company or team
- No public version number, changelog, or GitHub star count exists for this product as of this review
π In One Sentence
Cognira is a solo-developer AI chat assistant with a filtered "guarded learning loop," private per-user memory, and a local Windows/macOS/Linux binary (Cognira Entity) included on every pricing tier.
π¬ In Plain Terms
Instead of a chat app that forgets everything between sessions, Cognira saves what you correct or teach it to a private memory you can inspect and delete one item at a time, and also offers a downloadable version of the same models that the vendor says runs on your own computer instead of a server.
πNote: This review is the deep-dive companion to Cognira's entry in the Local LLM Software Directory β see that page for how Cognira compares at a glance to other local and hybrid AI tools.
What Is Cognira?
Cognira is a chat assistant built and operated by one independent developer, Thomas Conway, per cognira.dev's own "built by Thomas Conway β a solo, independent build, not a lab or a startup with a marketing team" statement. The site frames its core idea as a "guarded learning loop": the assistant adjusts based on a user's corrections and ratings, but changes are filtered for safety and formatting and checked before going live, rather than applying immediately from a single interaction. cognira.dev is itself a same-day rebrand β at the time of this review, the founder's prior product, Conway AI (conw.ai), was mid-migration to the cognira.dev domain, so some site content may still be settling.
- Core function: a chat assistant with a private, per-user memory and a filtered self-adjustment loop, available as hosted chat and as a downloadable local binary
- Underlying model: cognira.dev names it "Cognira Retrain 12B," described as proprietary β this review found no independent way to verify the model's architecture or training beyond the vendor's own description
- Hosting: hosted chat runs on a single 16 GB iMac and is served through MLX (Apple's on-device machine-learning framework), per cognira.dev
- Developer: Thomas Conway, an independent solo developer β not a company, lab, or funded startup
Key Features
Cognira's feature set centers on two things: a filtered learning process and a memory system a user can inspect and control. Both are described on cognira.dev as follows.
- Guarded learning loop β a user teaches the assistant through corrections or ratings; the site states that input is then filtered for unsafe or poorly formatted content, and that updates must pass checks before they go live, rather than a single chat being able to change the system's behavior directly
- Private, per-user memory β cognira.dev states that teaching it "a preference, a fact, or a word" saves that item to a private memory immediately, and that the next reply using that memory tells the user exactly which memory it drew on, with a one-tap "Forget" option
- Cognira Entity (local binary) β a sealed binary for Windows, macOS, and Linux, described by the vendor as running the same models as hosted chat on the user's own disk, with optional weight-level retraining, included on every pricing tier β see Local vs. Hosted below for what this claim does and does not establish
- Weekly token allowance β every tier (including the free tier) is metered by a token allowance that resets weekly on Mondays and counts both input and output, rather than a fixed monthly quota
Local vs. Hosted: What Cognira Entity Actually Is
cognira.dev makes two claims about Cognira Entity that are worth separating: what it says about where computation happens, and how it is packaged commercially. Both matter for anyone evaluating Cognira specifically as a *local* AI tool.
- cognira.dev states that for Cognira Entity, "inference, memory, and optional weight-level retraining all stay on your disk. Nothing leaves." Taken at face value, that is a genuine on-device inference claim, not just a downloadable app that still phones home
- At the same time, the pricing page lists Cognira Entity as a feature included inside the Free/Pro/Max tiers, which are otherwise defined entirely by a shared weekly token allowance with hosted chat β this pattern is more consistent with an account-gated, metered client than with a fully offline, unlimited-use local tool in the sense that Ollama, LM Studio, or GPT4All are
- cognira.dev/docs returned a 404 at the time of this review, so there is no independent technical documentation β such as an offline mode, a network-activity log, or a technical whitepaper β to confirm which of the two framings above is closer to how Cognira Entity actually behaves
- No GitHub repository exists to inspect the binary's actual behavior directly
Getting Started with Cognira
Cognira is used by signing up at cognira.dev β there is no separate installer page or download link independent of an account; every tier, including the free one, requires signing up first. Cognira Entity is then made available as a companion download once an account is active.
Path | What you get |
|---|---|
| Sign up at cognira.dev (Free tier) | Hosted chat, 25K tokens/week, Cognira Entity access |
| Pro (Β£15/month) | 500K tokens/week, same features as Free |
| Max (Β£30/month) | 2M tokens/week, same features as Free |
All three tiers include identical features β the only difference cognira.dev documents between them is the weekly token allowance. There is no separate one-time purchase or perpetual license; Cognira is subscription-only above the free tier.
Cognira Pricing and License
Cognira has a free tier and two paid subscription tiers, and is closed-source. Pricing is stated in GBP on cognira.dev/pricing; no LICENSE file or published EULA was found for the software itself.
- Free: Β£0, 25K tokens/week, includes the full Cognira Retrain 12B model (not a reduced "lite" version, per cognira.dev), private memory, and Cognira Entity
- Pro: Β£15/month, 500K tokens/week (20Γ the Free allowance) β same feature set as Free
- Max: Β£30/month, 2M tokens/week (4Γ Pro, 80Γ Free) β same feature set as Free
- Tokens reset weekly on Mondays and count both user input and model output
- License: proprietary/closed-source. No LICENSE file, EULA, or public source repository was found for Cognira or Cognira Entity as of this review
- No separate one-time or lifetime purchase option β Pro and Max are recurring monthly subscriptions
Who Should Use Cognira?
Whether Cognira fits depends on how much weight you put on an early-stage, single-founder product with no independent code review, against its private-memory and local-binary features.
Consider Cognira if
- You want a chat assistant that keeps an inspectable, deletable private memory of what you have taught it, rather than starting from zero every session
- You are comfortable with a proprietary, closed-source product from a single independent developer rather than an established company
- You want the option of a local binary (Cognira Entity) alongside hosted chat, and are comfortable with the ambiguity described in Local vs. Hosted above
- A free tier with 25K tokens/week is enough for casual, occasional use
Wait, or look elsewhere, if
- You need to independently verify how a tool handles your data β see What Is Cognira above for what this review could not confirm
- You need a tool with an established track record β Cognira and its cognira.dev domain launched the same day as this review, as a rebrand of the founder's prior product (Conway AI)
- You specifically need offline, unlimited, account-free local inference β this review could not confirm that Cognira Entity works that way; see Local vs. Hosted above
- You need an open-source or independently auditable license β Cognira is closed-source with no published LICENSE file
- You need voice input/output, plugins, or tool-calling β none were found on cognira.dev
Cognira vs. Other Personal AI Assistants
Cognira sits in the same broad category as other personal-memory-focused AI assistants. Here is how it compares to the closest alternatives β see the Local LLM Software Directory for the full catalog.
- Khoj β an open-source personal AI with long-term memory and document search that can run against local models; unlike Cognira, its source code is publicly auditable. See the Khoj review.
- Second Me β an open-source project for building a persistent, personal AI memory model, with a public GitHub repository Cognira does not have an equivalent of. See the Second Me review.
- Aori β a personal AI agent focused on task automation rather than Cognira's memory/chat focus, useful to compare if agent actions matter more to you than persistent memory. See the Aori review.
- Jarvis β a macOS-only personal assistant, narrower in platform scope than Cognira's cross-platform Entity binary but from a similarly small independent team. See the Jarvis review.
Common Mistakes When Evaluating Cognira
Most confusion about Cognira comes from conflating its two products, or from treating vendor marketing copy as independently verified fact.
Mistake 1: Assuming Cognira Entity is Apple-Silicon-only
Cognira Entity is described on cognira.dev/pricing as "a sealed local binary for Windows, macOS, and Linux" β cross-platform. Apple's MLX framework is used for the separate hosted chat service, which the vendor says runs on a single 16 GB iMac, not for the downloadable Entity binary specifically.
Mistake 2: Assuming "runs on your disk" means fully offline and unmetered
cognira.dev states Cognira Entity keeps inference and memory on-disk, but Entity is also listed as a feature bundled inside the same token-metered Free/Pro/Max tiers as hosted chat, and no independent technical documentation was found to confirm offline, account-free use. See Local vs. Hosted above.
Mistake 3: Treating cognira.dev as an established product with a track record
cognira.dev is a same-day rebrand of the founder's prior product, Conway AI (conw.ai), launched the same day as this review β see What Is Cognira above for what that means for its track record.
Frequently Asked Questions
What is Cognira?
Cognira (cognira.dev) is a chat assistant built by one independent developer, Thomas Conway, with a filtered "guarded learning loop" and a private, deletable per-user memory.
Is Cognira free?
There is a free tier (Β£0, 25K tokens/week). Paid tiers are Pro (Β£15/month, 500K tokens/week) and Max (Β£30/month, 2M tokens/week) β see Cognira Pricing and License above.
What license does Cognira use?
Proprietary/closed-source. No LICENSE file or public EULA was found for Cognira or Cognira Entity as of this review.
What is Cognira Entity?
A sealed local binary for Windows, macOS, and Linux, included on every pricing tier, that the vendor says runs the same models as hosted chat on the user's own disk. See Local vs. Hosted above for what this review could and could not confirm about that claim.
Is Cognira Entity limited to Apple Silicon Macs?
No. cognira.dev/pricing describes it as a binary for Windows, macOS, and Linux generally, not Apple-Silicon-specific. Apple's MLX framework applies to the separate hosted chat service.
Does Cognira Entity run fully offline?
The vendor states inference and memory stay on-disk, but Entity is also listed as a feature inside the same token-metered pricing tiers as hosted chat, and no independent technical documentation confirms offline, account-free use. This review treats that as unresolved rather than confirmed either way.
Does Cognira have a GitHub repository?
No. This review found no GitHub repository, star count, or public source code for Cognira or Cognira Entity.
What model does Cognira use?
cognira.dev names it "Cognira Retrain 12B" and describes it as proprietary. This review could not independently verify the model's architecture or training beyond the vendor's own description.
What is Cognira's current version?
No public version number, changelog, or release history was found for Cognira or Cognira Entity as of this review β verify directly with cognira.dev if a version matters to your decision.
How is Cognira different from Conway AI?
Cognira is a rebrand: cognira.dev replaced the founder's prior product, Conway AI (conw.ai), in a migration that was still in progress at the time this review was verified.