SMF·CLAUDE
Can a prompt replace Claude?
AI assistants — AI assistant
Exhibit tracking slip
Verdict
A Claude-shaped wrapper — chat over the Anthropic API, document upload, saved project contexts — is genuinely easy to build. What stays out of reach is the model itself, its long-context quality, and Artifacts and Projects as a polished product surface rather than a plain API call.
Exhibit A — The prompt
Received on30.07.2026Build a thin client on the Anthropic API: a chat interface with streaming replies, a conversation sidebar, and Markdown rendering. Let a conversation carry an editable system prompt, with Sonnet as the default model and the option to switch to a larger model per conversation. Add 'projects': a project groups related conversations and holds a handful of reference documents (plain text or PDF) that get prepended as context to every chat inside it — use the API's prompt caching on those documents so repeated context doesn't get re-billed every message. Store everything in a real database with full-text search across all conversations, and let any thread export to Markdown. Show a running cost estimate per conversation and per month from the API's usage data, so keeping this instead of paying a flat fee is a decision backed by a number, not a feeling.
Do not build Artifacts as an interactive canvas, the mobile apps, or flat-rate pricing for heavy use — if the cost meter shows genuinely heavy usage, a flat-rate plan wins on price, and the honest move is to switch back rather than keep optimizing around it.Opening prefills the prompt — press enter to run it.
Exhibit B — What you lose
- B.1 the frontier model and its continuous upgrades
- B.2 long-context performance tuned by Anthropic
- B.3 Artifacts as an interactive canvas, not just returned text
- B.4 the mobile apps
- B.5 flat-rate pricing that doesn't punish heavy use
Prior art
Exhibit C — Why people still pay: frontier model/platform
The quality sits in the model and the product surface around it, not in the chat window — a wrapper gets you the same API, but none of Anthropic's product work on top of it.
Questions
Can I import my Claude conversation history?
Claude.ai lets you export your data, including conversations, and the export maps reasonably well onto this build's storage — a short one-off script converts the export format.
Will it work on my phone?
As a responsive web page, yes, if it's hosted somewhere reachable. There's no native app and no offline mode.
What does it actually cost to run?
Hosting is a few dollars a month; token costs for light-to-medium chat use are often well under the flat subscription price, especially once prompt caching is wired up for repeated project context.
What's the one thing that genuinely doesn't survive the rebuild?
Artifacts as a real product surface. The API can return code or documents as text, but the interactive, versioned canvas Claude.ai renders them into is product work this wrapper doesn't replicate.
Related tools
Receipt