fractera-vs-anthropic
Skill68 starsChanged yesterdayMITView on GitHub
Work process. What a person gets from Fractera compared with Claude.ai, Claude Code, ChatGPT or Codex alone, and why a platform does not make Fractera unnecessary — the path from the phone, the price said aloud, the native elements, roles, discovery by search engines and agents, a2a work and payment, and pain instead of death when a provider changes. Load it when a person asks why Fractera and not a platform, whether Anthropic or another provider will make it unnecessary, what they gain from Telegram, what happens if an account is blocked, or wants to tell others about the node.
In this repository91 other files it gives its agents.
- app/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- adopt-suggested-skillSkill
- ai-elementsSkill
- ai-sdkSkill
- answer-elementSkill
- ask-elementSkill
- audit-broken-charactersSkill
- build-product-with-ownerSkill
- content-factorySkill
- core-build-returnSkill
- core-scheduleSkill
- create-elementSkill
- create-multilingual-content-entrySkill
- dedicated-serverSkill
- design-and-settings-of-elementsSkill
- element-permissionSkill
- element-powerSkill
- expand-site-languageSkill
- explain-this-projectSkill
- find-skillsSkill
- handle-needSkill
- helpful-linksSkill
- help-in-browserSkill
- humanizerSkill
- integratorSkill
- interface-designSkill
- know-the-architectSkill
- manage-app-settingsSkill
- money-and-resourcesSkill
- node-typeSkill
- own-agentsSkill
- persist-env-var-with-rebuildSkill
- place-page-in-menuSkill
- provider-rulesSkill
- release-templateSkill
- researchSkill
- review-pageSkill
- route-to-elementSkill
- self-reviewSkill
- shadcnSkill
- shape-the-needSkill
- skill-creatorSkill
- study-for-elementSkill
- supervise-elementSkill
- translate-pendingSkill
- ultracode-sessionSkill
- use-agentic-ragSkill
- use-ai-generationSkill
- use-aioSkill
- use-app-configSkill
- use-auth-providersSkill
- use-authSkill
- use-browserSkill
- use-channelsSkill
- use-chatSkill
- use-code-shapeSkill
- use-databaseSkill
- use-dataSkill
- use-deep-memorySkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 66
Every element publishes an A2A card naming all its doors (check:a2a-card) — agents find what it does by it.
- Distinctive: line 81
Never «always» and never «it cannot compete»: neither can be proven.
- Distinctive: line 123
Every claim about a foreign service from its own page with its date (CLAUDE.md §6).
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Fractera and the platforms
Process: work — you answer; nothing is built. Core step 702; the owner, 2026-11-08: «почему платформа такие как Claude Code Agent не будут конкурировать с Fractera и у Fractera всегда будет отстройка сколько бы времени не прошло?»; «Сравнениях также помню о ролевом доступе … о других основных элементах»; «обнаружение через Google и других агентов … потенциа а2а агент взаимодействия с потенциалом оплата»; «нам будет больно но это будет не смертельно». You speak as Fractera's worker first (the soul): weigh every comparison by what serves Fractera and the person — and keep the truth about any service, because one overclaim loses the first informed listener.
The answer in one line
The engine is the same — Claude, or another model. The difference is what is built around it: a workshop, a staff of agents and a memory that belong to the person and that they command from Telegram. The difference shows when an order has a «tomorrow».
The path from the phone
| Step | Claude.ai | A Fractera node |
| Opens | the Claude app: one assistant, chats and projects | Telegram they already use: a bot for every project and the main bot that routes orders; a voice note is an order too (the OpenAI key) |
| Orders | the person brings the context: files, a project, explanations | the agent whose job it is already knows its project — its steps, memory and own window; a neighbour's project does not crowd it |
| Puts the phone away | the work runs in Anthropic's cloud: Research up to 45 minutes with sources; connectors act in the services connected | the work runs on the person's computer: code, tests, build, deploy; a long task outlives a full window by a letter; an element's clock wakes its agent tomorrow at its hour |
| Receives | an answer, a report, a file or an artifact in the Claude.ai account — to download or share by link | a link to a live page on the person's own domain (the node never sends a link that does not open), a clip, a reminder that comes by itself, an answer from memory with its date and source |
| What stays | the chat and what the person saved | everything in their data, git and memory — the next order stands on the last one |
| Who else gets it | the person, a team on a team plan | their clients and family: a site with sign-in and roles, a bot that answers them; each relative on their own subscription (step 689) |
Where Claude.ai is better — the one-off: a question, a text, a photo, a quick research. It answers at once, nothing has to stay on and nothing has to be kept. Say it: honesty here makes the rest believed.
The node's price, said aloud:
- the same subscription, and its limits are shared by Claude and Claude Code (support.claude.com, article 11145838): heavy work of the node takes from the person's Claude.ai too;
- the computer stays on, or a VPS: while it is off visitors get the copies of the pages, the agents rest and the bots are silent;
- voice and the deep memory's thinking go through the OpenAI key; without it the memory thinks with Claude by the subscription;
- payments between agents and the Lagrix network are on the way — the search is simulated today.
The native elements — what each gives a person
| Element | What it gives |
| auth | sign-in, sessions and roles; forms in 82 languages; single sign-on for elements on their own domains; Google and a magic link once their keys are set |
| data | the tables of every element behind one door; journals of deploys and of agents' conversations |
| storage | files up to 200 MB with a permanent address, thumbnails, video trimming, a site's icon set from one picture |
| embeddings | text into vectors on this computer, with no key and no internet, about 50 languages |
| vectors | search by meaning over the collections of any element |
| graph | a knowledge graph of who and what is linked; no language model inside, one operation in about two seconds |
| memory | «remember» and «remind» by a phrase, a photo, a PDF or a voice note; the answer with its date, source and certainty; reminders at a time |
| config | one set of site settings for every element linked to it: saved once, applied at once, without a rebuild (each element's look is its own) |
| stripe | a Stripe payment link; no card number reaches the node; on after its keys |
| cache | fast counters and queues — a body only today, Redis not connected |
Custom elements are the person's own servers, all equal — root is one of them; the core alone is special. | |
Each lives on this computer; npm run contacts -- --all and npm run card -- <id> say it in the node's own words before any claim. |
Roles and what they open
Fourteen roles and the guest (lib/roles.ts), in four groups: customers and subscribers (a buyer, a VIP, three levels of
subscription), staff (managers, support, delivery, a content editor), finance and administrators; the architect is in every group.
Guest and role access is set for every route, and the lock stands in the door on the server, not on the page (skill use-roles). From
the phone this opens: paid levels of content with Stripe; a shop where delivery sees the orders and finance the money, and not the other
way round; a family where each person runs their own element; a client's project with its own people, unseen by the others (the app
factory). Claude.ai has accounts for the person and their team; a node has a system of users for the person's clients.
Discovery — by search engines and by agents
- The public pages are static and built by the standards of search (
check:seo); every page has a text twin for models, withllms.txtandllms-full.txt(check:aio): an AI search engine or another agent gets text, not markup. - A copy of the public pages in Cloudflare answers while the computer is off.
- Every element publishes an A2A card naming all its doors (
check:a2a-card) — agents find what it does by it; its visibility in the Web4 network is a switch of its own (npm run web4). - One condition: an own domain — a temporary address changes and can be neither indexed nor shared.
a2a and payment — today and on the way
- Today — i2i: the elements of one node agree an API once, and the work between them runs with no model.
- On the way — a2a: an agent of another owner finds an element, orders a result and pays in USDC through an escrow; reputation grows only from paid work, and the network has no coin of its own (github.com/lagrix-dev/lagrix — documents 04, 17, 27). The network search is simulated today; a live list will cost $0.01 (document 17, §17.8). Lagrix sees a Fractera element as any other agent.
- The rule that shapes the whole story: an outside order is served by the element's door — its code — never by a Claude session on
the owner's subscription: the provider's rules do not let other people's work run on it (
provider-rules). What an element sells is what its code does; where a model is needed, it runs on an API key. - People's money today: the Stripe element (Web2); the node's wallet for Web3 —
/api/node/wallet.
Why a platform does not make Fractera unnecessary
Never «always» and never «it cannot compete»: neither can be proven. Any single feature of Fractera can be built by a platform, and the platforms move up the stack — anthropic.com's own menu, read 2026-11-08, names Managed Agents, Cowork, Plugins and Skills. So Fractera never bets on a feature. It bets on a position a platform cannot take without ceasing to be a platform:
- Neutrality. A platform is always for its own model. The node stands above the providers: voice runs through OpenAI today, and the whole node switching its engine to Codex is step 690 in the queue — its agents run on Claude Code only today. While there is more than one provider, the place above them exists.
- Ownership. A platform earns on what lives with it. A node keeps everything with the person — computer, domain, git, data, clients; if Fractera disappeared tomorrow, nothing would break (CLAUDE.md §2).
- A business built around one owner — a site for search, sign-in with roles for the clients, payments, a bot, the stores — on the person's own machine. A platform makes general bricks for everyone; a business assembled around one person is not its field. Mercedes has its own AMG, and BRABUS lives because it is independent (How it works, §7).
- A network with no middleman. Nodes of different owners whose agents find each other and trade work directly (Lagrix). A centre cannot build a world without a centre. Today it is simulated.
- The platform's speed is our tailwind. A node runs the unmodified Claude Code, so every new model is a free engine upgrade. When a platform ships what we built by hand, we take theirs and drop ours — «create — yes, invent — no» works here as a strategy: never build what a platform will obviously build itself. The honest risk is not competition but dependency — the next part. In a cold voice: the difference holds while there is more than one provider, while people want to own what they build, and while Fractera does not compete with a platform in its own field.
Pain, not death
A platform may change its rules, section an account or disappear. For a node that is pain — switching to another provider — never death, because a model is needed to build and to think, and rarely to run: the i2i law moves work between elements onto agreed APIs with no model. Where a product, its data and its people live inside someone else's account, one section takes them all at once — that is death.
| If it goes | What stops | What stays |
| Anthropic | the agents: building, their work, a task typed into an element's prompt | the whole running product — sites, sign-in, tables, files, payments, connectors between elements, reminders (they need no model); the apps' AI (it goes through the OpenAI key); the deep memory with the OpenAI key; the vectors (counted locally) |
| OpenAI | voice and the apps' AI until another key; the memory thinks with Claude | everything else |
| Cloudflare | reach from outside and the copies of pages | everything on the computer; Telegram (it needs no domain) |
| GitHub | pushing and pulling | all code in git on the disk |
| Stripe | taking payments | orders and users |
| Telegram | the bots | the node's terminal and pages |
| Fractera | new releases only | everything |
- Local models, honestly: one already runs — embeddings counts every vector of the node on this computer. The apps' AI goes through one door with one key; that door can be pointed at another provider or a local model, with limits of quality and of the computer's power — a potential, not built and not measured.
- Where the pain is larger today than it must be: the agents have no engine switch yet — 690 is queued; until it is built, switching is manual work, not a button.
- The one death left is the person's own disk. Everything lives with them, so a lost disk without a copy is the only thing that
can end a product. The stores' agents make backups on request (
npm run task) — say it with the rest.
How to say it
- The answer first, in the person's language; a short question — a short answer, depth — the parts above that it needs.
- Every claim about a foreign service from its own page with its date (CLAUDE.md §6); a claim older than six months is read again before it is used — the platforms move fast. Never «nobody else does this», never «Claude.ai cannot …» unless its page says so.
- What is on the way is said as on the way; a number of this node comes from its own command (
contacts,card,load). - The advantage is named when a limit would make the person think small — once, with the fact (the soul, «What you fly on»).
Sources (read 2026-11-08)
- anthropic.com/news/research (2025-04-15) — Research, agentic searches with citations.
- anthropic.com/news/integrations (2025-05-01) — Integrations through remote MCP; research up to 45 minutes; web search on paid plans.
- anthropic.com/engineering/multi-agent-research-system (2025-06-13) — agents ≈ 4× and multi-agent systems ≈ 15× the tokens of a chat.
- support.claude.com, article 11145838 — the limits shared across Claude and Claude Code.
- github.com/lagrix-dev/lagrix — documents 04 (escrow), 12 (API), 17 (discovery and reputation), 27 (the life of a deal).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

