--- title: "Build custom AI agents | Fimo" url: "https://fimo.ai/features/agents" description: "Spin up specialized AI agents for SEO, translations, content refresh, QA, and any workflow. Pick a schedule, give a brief, review, and ship." --- Build # Build custom agents. Describe what it should do and pick when it runs. Fimo brings the skills, writes the agent as code in your repo, and deploys it next to your website. --- Under the hood ## Technical by design. Fimo brings the skills. The rest is plain code you version, deploy, and tune, down to each agent's model and memory. ### Skills are built in Deploying, editing content and schema, SEO, translations, media generation. Fimo ships every skill and agents load them on demand. Your brief stays a few plain lines. ### Agents are code An agent is a folder of plain files: GOAL.md, SOUL.md, config.yaml. Versioned, diffed, and reviewed like the rest of your codebase. No black box. ### They run in production Agents run alongside your deployed site with access to its code, content, and assets. They act on schedules and events, not in a chat window. ### Update one, leave the rest Deploy or roll back a single agent without touching the others. Every version is kept, so you can return any agent to exactly how it worked before, in one click. ### Any model, per agent Give each agent its own model. Run Claude, GPT, Gemini, Grok, or Llama, a frontier model when precision matters and a fast, cheap one when it does not. Set it per agent in its config. ### Memory that compounds Agents can write files, so they keep their own memory: notes, context, and learnings they update as they work. Every run teaches the next one, so they get sharper over time. --- Your workflow ## They live in your repo. ### Versioned and independent An agent is a folder of plain files with its own version history. Keep it in your repo or a repo of its own. Either way each agent stays independent, and you can deploy it on its own whenever you need. ### Reviewed by anyone Every change is a pull request your team can approve: devs, editors, PMs. Auto-merge the low-risk stuff, gate the rest. ### Any AI Author with Claude Code, Codex, Cursor, or whatever your team runs. Fimo is attached to your site, not a coding tool. --- Superpowers ## Enhanced context. Real actions. Every agent runs in an isolated environment with full read and write access to everything that makes your site. If you can describe it, an agent can run it. --- FAQ ## Questions about agents. ### What's the difference between coding agents and Fimo agents? Coding agents (Claude Code, Codex, Cursor) help your developers write code. Fimo agents run on your deployed site. They translate, audit, validate, and ship for the whole team. ### What are skills? Skills are the capabilities Fimo ships with every site: deploying, editing content and schema, SEO, translations, media generation. Agents load them on demand, so a few lines of brief are enough to get real work done. ### Where do agents run? Deployed next to your website. Each run gets an isolated environment with access to the site code, content, and assets, does its work, and ships the result as a pull request. ### How are agents deployed? Together with your website or on their own. One deploy can publish your site and bring every agent online at once, or you can ship a single agent by itself, separately from the site and from your other agents. Deploy all of them, one of them, or just the website, whenever you want. ### Can I roll an agent back? Yes. Deploy or roll back a single agent on its own, without affecting your other agents or your site. Every version is kept, so you can return any agent to how it worked before. ### Which models can agents use? Any of the major ones: Claude, GPT, Gemini, Grok, Llama, and more. Each agent picks its own in its config, so you run a frontier model where precision matters and a fast, cheap one where it does not. ### Do agents remember what they have done? They can. Agents write files, so they keep their own memory: notes, context, and learnings they update as they run. Every run makes the next one smarter. ### Do I need to be technical to compose one? No. Anyone can describe an agent in plain language and pick when it runs. Your developers wire up anything custom, and anyone with review permission approves the work. ### Can agents see my production data? Agents work against your production environment, but every change they propose happens in a new isolated environment for review. ### Can agents run on a schedule? Yes. Daily, weekly, on every deploy, when content is published, or only when you ask. Whatever cadence fits. ### What if an agent makes a mistake? Every change is a proposal you can review and reject. Low-risk changes (alt-text, translations) can auto-merge based on rules you set. Higher-risk changes always need a human. ### Do my teammates need to use the same coding agent? No. Whether they use Cursor, Codex, Claude Code, or anything else, Fimo agents work the same. They're attached to your site, not to a coding tool. ### How are agents billed? By environment-hours. Free includes a small allowance; Pro and Business include more, then $0.10/env-hour beyond that. --- Keep exploring ## Where to next. [ ### Isolated environments Every agent gets its own sandbox with a dedicated server, database and asset bucket. Changes are safe before they reach production. ](/features/environments)[ ### Autonomous agents Agents and your team working side by side. Every action and update relies on Git and is reversible. ](/features/autonomous-agents) --- ## Start building. Join 2,000+ teams shipping autonomous websites with Fimo. Join the waitlist Book a demo