Fimo

Positioning: the layer after vibe-coding

Fimo is one of the more interesting products I have examined this year, because it is built on a specific observation about where AI website creation actually breaks. Tools like Claude Code, Codex, Cursor and Lovable can generate a working site in hours. What they do not give you is everything that happens after launch: a way for non-developers to edit content, a way to keep pages from going stale, and any systematic work on SEO and AI-search visibility. Fimo, founded in 2026 by Pierre Burgy and Aurélien Georget — the founders of Strapi, the open-source headless CMS — positions itself as that missing layer. Not a website builder, not a coding agent, but an "autonomous website platform": an AI CMS that sits on top of code your coding agent produced, plus a system of agents that keep working on the site after you ship it.

The pitch is genuinely different from the CMS incumbents it names on its comparison pages. Where Contentful or Webflow charge for sites, workspaces, seats and add-ons, Fimo says it meters the work itself: you pay for what agents do, not for logins. Where AI builders generate a site and stop, Fimo's agents are supposed to run on schedules and events forever — auditing SEO Monday mornings, translating releases, refreshing stale content — with every change landing as a pull request you review. It is a bet that the unit of value in web publishing is shifting from "storage and seats" to "work actually done", and it is a bet made by people who spent a decade watching CMS buying behaviour up close.

What it actually does

Fimo attaches to an existing JavaScript project — typically output from Claude Code, Codex or Cursor — via a setup command (npx fimo@latest init) that maps the project and turns its content into a CMS. Non-technical teammates then edit text, links, images and video directly on the page, in structured fields, or by chat, and every edit lands as a clean commit in the repository. Design-level changes are prompted rather than dragged (ask the chat to change a button color) or routed through a design tool like Figma. The website itself stays real framework code; Fimo is explicit on its own AI-facing documentation that it is not a no-code builder and not a proprietary page format.

The agents are the differentiating half. Each agent is a folder of plain files in your repo — GOAL.md, SOUL.md, a config — with a goal, a schedule, and permissions. Fimo ships built-in skills (deploying, content and schema editing, SEO, translations, media generation), and agents load them on demand. Each agent can pick its own model (Claude, GPT, Gemini, Grok, Llama), keeps file-based memory that persists across runs, and works in an isolated environment: every agent run and every branch gets a disposable copy of the stack — server, database, asset bucket — so production is untouched until a change merges. Permissions decide what ships alone: low-risk changes can auto-merge, the rest stop and ask via the Fimo inbox, Slack, Teams or Linear. One command deploys the site and its agents together to managed edge hosting with SSL, CDN, custom domains, preview URLs and instant rollbacks. There is also SEO/GEO plumbing — clean Markdown rendering per page, JSON-LD generated from CMS content, server-side rendering — and an SEO agent that audits pages and proposes fixes.

Pricing, at time of writing (fimo.ai/pricing), is one plan per organization with hosting included and no per-seat pricing. Free: $0, 5,000 AI credits per month, 1 agent-hour, 5 GB bandwidth, 1 GB media storage, up to 3 sites and 3 members. Pay as you go: $0 monthly plus metered usage — overage at $2.00 per 1,000 credits, $0.75 per agent-hour, $0.15 per GB, with unlimited sites and members, custom domains and event-triggered runs. Pro: $149/month for the organization, 35,000 credits, 40 agent-hours, 200 GB, lower overage rates ($1.00 per 1,000 credits, $0.33 per agent-hour), 90-day version history, 1-year analytics, a 99.9% uptime SLA and priority support. Credits are straightforward: 1,000 credits ≈ $1 of AI hosted by Fimo, covering agent runs, Studio edits and image generation; building the site with your own coding agent never consumes credits.

My testing notes

Two concrete observations from working through the product surface.

First, the pricing arithmetic holds up better than most usage-based models I evaluate. I worked the metering numbers through: on the free plan, a typical agent run (Fimo says a run holds its sandbox for roughly 30 minutes) consumes about half an agent-hour, so the free tier gives you roughly two agent runs a month before the runtime allowance pauses — enough to see the loop work, not enough to operate on. Pro's 40 agent-hours is on the order of 80 runs a month at that same 30-minute figure, which for a site running four weekly agents is comfortable headroom. The "500 KB per visit" bandwidth estimate in their FAQ is openly a forecast, not a quote, and the page says so. That is the kind of honesty I rarely see on usage-metered pricing pages, and it made the cost model legible enough for me to model a real workload against it.

Second, there is a real tension between the "your code stays yours" positioning and the current infrastructure. Fimo's own AI-facing documentation states that code lives "in our private Git repository for now; GitHub and GitLab will be supported soon." So while the ownership claim is honest at the level of format — it is standard framework code with real git history, not an export button's worth of proprietary markup — day one of onboarding means moving into Fimo's repository hosting. I would treat that as the single most important operational caveat right now: the lock-in is low in principle and moderate in practice until GitHub/GitLab support ships. Relatedly, I noticed the homepage's "what developers say" block quotes Vercel's CEO, a Databricks PM and Blaxel's CEO; those are vendor-published testimonials and I could not independently verify them, so I read them as marketing, not evidence.

A third observation worth a line: the press coverage categorizes Fimo differently than Fimo categorizes itself. CMS Critic's launch write-up files it under "no-code website builder", while Fimo's own docs insist it is neither no-code nor a builder. Neither is lying — the CMS Critic author used it to generate and restyle a site conversationally, which looks like a builder from the outside — but the category confusion is a warning that Fimo is hard to evaluate against the tools buyers already know. It is genuinely neither Webflow nor Claude Code; it is what runs after both.

Limits

It is young. Fimo launched in 2026, and at time of writing the independent review corpus is essentially one positive CMS Critic hands-on and a Product Hunt listing with a single 5.0 review. I could not find long-term reliability data, uptime history, or third-party benchmarks. There is no free-form design canvas — visual editing covers content; structural design changes must be prompted or made in code. There is no traditional CMS import path from WordPress or Contentful documented yet, so migrating an existing large site means engineering work rather than a wizard. Multi-brand or multi-tenant agency setups are addressed in positioning ("agencies and studios") but I found no dedicated agency program or white-label documentation.

The agent model also shifts risk onto your review discipline. Auto-merge is a permission you grant; if you grant it loosely across SEO and content agents, the "clean commit history" becomes a stream of machine-authored changes nobody read. Fimo gives you the guardrails (PR review, per-agent permissions, per-agent rollback), but it does not make curation free. And because agent runs consume both credits and agent-hours, an over-eager agent on a large site can burn budget — the metering is legible, but it is still metering.

Finally: this is a single-vendor SaaS built by the Strapi team through a new legal entity (Strapi, Inc.), with pricing promises (99.9% SLA) that only bind on the $149/month tier. Startup risk is real, mitigated somewhat by the code-ownership story — if Fimo disappeared, your site is still standard code in a repository.

Verdict

Fimo rated 3.7 out of 5 on my rubric: the core functionality is ambitious and unusually coherent for a first-year product (weighted 40% — the agent-on-a-real-repo model is the deepest thing in this category I have tested), value is strong because metering is transparent and there are no seat fees (20%), usability gets credit from both CMS Critic's hands-on and the clean onboarding path (20%), but the ecosystem is nearly empty of third-party proof (10%) and reliability signals are a promise rather than a track record (10%). If you are shipping AI-built sites and care about owning the code, Fimo is the first product I have seen that treats "who runs this website on Tuesday morning" as the actual problem. I would pilot it on a real but non-critical site now, keep GitHub-hosted repos as the requirement once supported, and re-evaluate in a year when there is an uptime history to judge.

Sources

• https://fimo.ai/
• https://fimo.ai/pricing
• https://fimo.ai/ai-info
• https://fimo.ai/features/agents
• https://fimo.ai/features/visual-editor
• https://fimo.ai/features/environments
• https://fimo.ai/features/seo-geo
• https://fimo.ai/compare/webflow
• https://fimo.ai/compare/lovable
• https://cmscritic.com/vibe-coding-fomo-fimoai-offers-easy-website-building-with-ai-native-cms-capabilities
• https://www.producthunt.com/products/fimo-2