Grain vs Charming

Quick answer

Both keep what an agent built open after the chat ends. Charming does it with a hosted app runtime. Your AI writes a small web app through Charming's MCP server, HTTP API or CLI, and Charming runs it with a key-value store, file storage, per-person roles and scheduled Routines. Grain does it with a workspace of files. Your agent writes files, Grain publishes them at a URL, and an app can add Postgres, logins and scheduled jobs. Choose Charming for quick single-purpose apps that feel native inside ChatGPT and Claude. Choose Grain when the work should outlive any one host as files, or when you need a SQL database or open source code.

Fit first

What is different, verifiably

Where Charming is better

• Nothing to install. You connect Charming's remote MCP server in ChatGPT or Claude and ask for an app, and the app can render inline in that chat.
• The AI can call a running app's operations as tools, so the same assistant that built a tracker can read from it and write to it later.
• Per-person roles on each app. A viewer reads, an end-user can write data, and a collaborator can change the code, all enforced on the server.
• Secrets stay out of the code. The owner stores an API key once, and Charming keeps it away from app code and the model.
• Lower entry price. Pro is $8 a month, and Team is $14 per seat with a three-seat minimum.

Where Grain is better

• The work stays as files. You pull, checkpoint and restore every artifact with the CLI, instead of reading data back through each app's own operations.
• Postgres behind the URL. An app that outgrows a key-value store can use SQL tables, logins and scheduled jobs.
• The source is open under AGPL-3.0 on GitHub, so you can read how artifacts are stored and shared.
• Static pages publish the same way as apps. A write-up, a PR walkthrough and a dashboard each get a URL.
• Links can require a password, and the agent can read reader comments and act on them.

Moving between them

1. From Charming to Grain: ask your agent to call get_app_source for each app, or fetch it from the source endpoint. You need owner or collaborator access.
2. Export the data through the app's own operations. Charming documents no bulk data export on Free or Pro, and template copies leave data and secrets behind.
3. Run grain init in a folder with the exported code, then ask the agent to replace Charming's storage calls with files or the workspace's Postgres database.
4. Run grain push, grain checkpoint -m "import from Charming" and grain share --public, or --password <pw> for a private link. Saved versions, activity history, roles, Routines and stored secrets do not carry over.
5. From Grain to Charming: run grain pull, then ask your AI to rebuild the app on Charming with its key-value store. Postgres tables need reshaping into key-value data, and Grain comments, share links and custom domains do not carry over.

Choice rule

Choose Charming

if you want a small app with built-in storage and per-person roles, made and used inside ChatGPT or Claude, with nothing to install.

Choose Grain

if your coding agent's work should stay as files you control, an app needs Postgres, or you want open source code.

Use both

if quick personal tools live on Charming and the team's reports, dashboards and larger apps live in Grain.

Frequently asked questions