Claude, ChatGPT, and coding agents generate working apps in minutes — that forget everything. FormLM's Data API documents itself in Markdown, so agents integrate it on their own. No glue code. No server.
Ask Claude to build a habit tracker, a quiz, a feedback form, or a mini CRM, and you'll get a polished, working UI in minutes. Then you close the tab and the data is gone. Artifacts run sandboxed with no database and no persistence between sessions. Pages the agent writes to your disk can render beautifully from file:// — and still have nowhere to POST. The one thing agents can't generate for you is a place for the data to live.
The fix isn't "deploy a backend." It's an agent-readable data API: an endpoint simple enough that the agent itself can discover how to use it. That's the design constraint FormLM's Data API was built around — every endpoint answers ?help with a plain Markdown document, because Markdown is the one format every LLM reads natively.
Describe what you want to collect in FormLM — one sentence is enough, AI generates the form. Enable the Data API on the Publish tab and copy the endpoint URL.
Paste your endpoint with ?help into the agent's chat. It receives the field list, request format, and examples as Markdown. This is the whole integration prompt:
It POSTs JSON to the endpoint whenever it should record something. Submissions become structured records — visible in FormLM, exportable as CSV, queryable over HTTP.
# One message. That's the entire setup. "Log my daily focus sessions to this API: https://formlm.me/api/v3/share/YOUR_TOKEN?help Read the docs at that URL first, then submit today's session data whenever I tell you what I did." # Live example — open it yourself: https://formlm.me/api/v3/share/2Pe6j0r88k8InlC49EcqffnTnURYYgciMn2n4Fp0iozMiw?help
Serving a YAML spec or a Swagger UI helps developers, not agents. The ?help doc is written to be consumed in a context window: short, explicit, with copy-pasteable examples. Agents get it right on the first request.
Auto-create fields mean the first submission defines the columns, and new keys later just add new fields (up to 50). When your agent's app grows from "mood" to "mood + sleep + steps," the storage keeps up without a migration step.
Agents can pull data back: the query endpoint pages through records and summary returns aggregate stats — both with their own ?help docs. Log all week, then ask the agent to analyze the trend it recorded.
A plain HTTP endpoint works with function calling, MCP servers, automations, and cron jobs alike. For heavier workflows, FormLM also ships a CLI with MCP support that can generate entire assessment apps.
Habit, mood, and focus logs from a local page the agent wrote. The data accumulates across sessions — the thing artifacts fundamentally can't do.
Quizzes, calculators, and generators built in one chat session that POST each result as a structured record — so you can see usage and outcomes later.
An agent that runs on a schedule and logs what it did — research digests, price checks, file scans. The query API turns the log into its own memory.
Collect responses from an AI-built front end first; when it's time for scoring, bands, and personalized PDF reports, the same data is already in FormLM's engine.
Not on their own — chat assistants and artifacts run in a sandbox with no database and no persistence between sessions. But they can call an external API. Give the agent a FormLM Data API endpoint and it can POST data there on your behalf; the records persist on FormLM's side.
You don't write integration instructions. Appending ?help to any endpoint URL returns a plain Markdown document: the field list, the request format, and copy-paste examples. Paste that one URL into the agent's chat and it has everything it needs to submit valid data.
Nothing breaks. With auto-create fields enabled, unknown keys in a submission are added to the form automatically (up to 50 fields). An app that evolves from logging 'mood' to also logging 'sleep' and 'steps' just sends the new keys — the storage adapts.
Yes. The query endpoint pages through stored records and the summary endpoint returns aggregate statistics — both also serve ?help Markdown docs. A common pattern: an agent logs data all week, then you ask it to pull the query endpoint and analyze the trend.
Yes. The API is CORS-enabled, so a page the agent wrote that runs from your disk (file://), a sandboxed artifact, or a static host can POST directly. No server, no proxy, no deploy.
It's complementary. MCP and function calling are how an agent reaches tools; the Data API is a tool worth reaching for — a plain HTTP endpoint with self-describing docs. FormLM also ships a CLI with MCP support for richer agent workflows like generating full assessment apps.
Free endpoint, Markdown docs your agent reads by itself, storage that adapts.
Create your free endpoint →