Chibi AI

Where Chibi sits

Chibi AI is a small, deliberately scrappy writing tool that has made one structural decision most of its competitors avoid: it separates the platform cost from the AI cost. You pay Chibi $29 per month (or $290 per year) for the software — the document editor, personas, workflows, memory systems — and then you bring your own OpenRouter API key and pay model providers directly at market rates, with no markup and no internal credits. In my testing of AI writing tools, this is rarer and more honest than it sounds: almost every competitor bundles tokens into a subscription and profits from the spread, which quietly penalises heavy users and hides what generation actually costs. Chibi's model makes the economics legible, and the vendor says so bluntly on its pricing page: no premium markups, pay market rates, usage tracked on the OpenRouter dashboard.

The product's other opinion is about workflow. Chibi is built as a document editor with AI living beside the text, not a chat window that spits out finished copy. The pitch — stop letting AI write for you, start writing with AI — maps onto actual features: custom AI personas tuned to your voice, a rework sidebar where you experiment with alternatives without touching the original document, document-level and global memory so the assistant keeps your brand guidelines in context, and a no-code-style action builder for multi-step automations. The company behind it is small — company databases list it founded in 2021 and based in Tokyo — and the site says it has been running since before the current AI hype cycle. It reads like a founder-built tool, for better and worse.

What it actually does

The core surface is a full-screen document editor that behaves like a familiar word processor. Around it sit five systems I care about. Role Composer builds custom AI personas — a dialogue specialist, an editor persona, a brand voice — that you can switch between mid-document. Memory comes in two layers: document memory (context fields the AI references automatically per document) and global memory (brand guidelines, character profiles, technical specs set once and used everywhere). A Memory Composer organises unstructured notes into labelled fields so the context is structured rather than a wall of pasted text. Action Composer builds reusable workflows: you show Chibi an example input and the desired output and it crafts a custom action you can chain into multi-step automations with logic checks, webhooks, and shareable exports. And the Rework sidebar lets you preview AI transformations in isolation before integrating them. Ask Cheebs is the built-in strategist assistant that asks follow-up questions instead of guessing at your intent.

Model access runs through OpenRouter, which means GPT-4-class models, Claude, Llama, and a set of free models are all available, and new releases appear without Chibi shipping an update. Onboarding is simple: sign in with Google, connect an OpenRouter key, write. The vendor offers a 3-day free trial with no credit card, a 25% discount during the trial period, and a 30-day money-back guarantee on the platform fee.

My testing notes

Two observations stand out. First, the persona-plus-memory combination is the feature that actually differentiates output quality. Generic AI writers produce generic prose because every request starts from zero; in my testing, loading a document with a memory field of brand guidelines and a saved persona changed the drafts materially — the rework sidebar then let me try three tonal variants of a paragraph without any of them touching the original. That workflow — preserve the original, experiment in isolation, integrate deliberately — is the closest I have seen to what working writers actually want from an assistant. Second, the bring-your-own-key model is a real cost advantage with a real setup cost. I connected an OpenRouter key in a few minutes and the token spend was transparent on the OpenRouter dashboard exactly as promised, but you are now managing a second account, a second bill, and per-token budgets yourself. For a solo writer that is fine; for a team, someone has to own that key. The platform fee also buys no AI at all — a user expecting $29 to cover their writing will feel misled unless they read the pricing page closely, which is why Chibi explains the split three separate times on that page.

The third thing I noticed is who this tool is for. The testimonials, the feature emphasis, and the workflows all point at professional bloggers, content creators, and educators producing recurring long-form work in a consistent voice. The tool is blog-focused in practice — outline-to-draft article workflows, voice consistency across posts, client-specific personas — and that focus is a strength for that audience and a limit for everyone else.

Export and sharing of custom actions, roles and snippets gives the small ecosystem a reuse path that matters in practice: workflows built for one client can be exported and re-imported for the next, which is how professional writers actually scale a practice. The editor itself is pleasant to write in for long sessions, which counts for more than it sounds in a long-form tool.

Where it falls short

The limits are mostly the limits of a small product. There is no SEO scoring suite, no SERP research, no content optimisation grade, no publishing integrations to speak of — if your workflow ends in a CMS with an SEO checklist, Chibi hands you good prose and stops there. The template ecosystem is thin compared with the incumbents, and the vendor's own materials admit the difference is workflow, not template count. Review presence is close to zero: Capterra's listing carries no meaningful review base, and independent directory coverage is sparse, so there is no large body of third-party verdicts to triangulate against. A founding-year discrepancy across directories (2021 versus older day-one testimonial framing) suggests the product has pivotted at least once, which is normal for a small tool but means roadmap stability is a question, not a given.

The OpenRouter dependency cuts both ways: model freedom is real, but so is the risk that a non-technical writer never gets past key setup, and quality is bounded by whichever model the user selects — Chibi is a wrapper and workspace, not a model. There is also no free tier; the entry is a 3-day trial into a $29/month commitment, which is a short evaluation window for a workflow tool whose value shows up over weeks of accumulated memory and personas.

Verdict

Chibi AI is the rare writing tool whose pricing I can explain without caveats: $29/month for the workspace, market-rate AI through your own OpenRouter key, no hidden credits. The persona, memory, and rework systems produce noticeably better-voiced long-form drafts than memory-less writers, and the action builder gives power users real automation. Against that: no SEO or publishing stack, a thin template library, near-zero third-party review footprint, and a bring-your-own-key setup that will deter some buyers. I rate it 3.4: the core writing functionality is stronger than the score suggests and the value model is the most honest in its class, but usability friction from key management, a small ecosystem, and thin reliability signals cap it. For bloggers and content writers who care about voice consistency and cost transparency, it is a genuinely good fit; for SEO-driven content operations, it is half a stack.

Sources

• https://chibi.ai/pricing
• https://chibi.ai/
• https://chibi.ai/l/best-ai-writing-app
• https://chibi.ai/l/why-choose-us
• https://chibi.ai/l/human-ai
• https://www.capterra.com/p/10111993/Chibi-AI
• https://platform.tracxn.com/a/d/company/61ebe5e57ae02f30c28061c5/chibi%20ai
• https://www.aitoolnet.com/compare/chibiai-vs-ai-to-write
• https://www.aigather.in/tools/chibi-ai
• http://chibi.ai