Dropstone

What is Dropstone?

Dropstone is an AI coding agent from research lab Blankline that keeps one long-term memory across chat, the CLI, the SDK, and your editor. It writes and ships code on open-weight models with a one-million-token context window, and can also watch an inbox, place phone calls, and control smart home devices.

Top Features:

• Shared memory: lessons taught in the CLI carry over to chat and calls.
• Proactive work: keeps running between sessions and acts on what it already knows.
• Approval gates: every tool call and real-world action waits for your sign-off.

Use Cases:

• Large refactors: load a whole codebase and plan changes before applying them.
• Inbox monitoring: have the agent watch your email inbox around the clock.
• Phone calls: let it place real calls on your behalf after approval.

Who Can Use Dropstone?

• Engineers: get a terminal coding agent that remembers project rules between sessions.
• Teams: self-host the stack and keep prompts and code out of training.
• Founders: hand off inbox watching and routine calls alongside daily coding work.

Pricing

• Free ($0): limited monthly usage of the Fast and Pro models via CLI.
• Pro ($15 per month): more usage, the Heavy model, and smart home device control.
• Max (from $75 per month): five or ten times Pro usage, priority routing, and phone calls.

Pros and Cons

Pros:

• Persistent memory: teach preferences once instead of repeating them in every session.
• Low cost: a free tier and fifteen dollar plan keep experimenting cheap.
• Privacy focus: zero data retention and no training on customer prompts or code.

Cons:

• Broad scope: mixing coding, calls, and smart home control can feel unfocused.
• Tier locks: phone calls need Max, and smart home control needs Pro.
• Rotating models: the open-weight models can change monthly, so results may shift.

FAQs:

1) Is there a free plan?

Yes, the Free plan costs nothing and includes limited monthly usage.

2) Which editors does it work with?

Extensions exist for VS Code, Cursor, and Windsurf, plus a terminal CLI.

3) Can it act without being prompted?

Yes, it runs proactively, but every action waits behind an approval gate.

4) Does it train on my code?

No, prompts, code, and outputs are never retained or used for training.

5) How large is the context window?

Every tier gets a one-million-token window that can hold a whole repository.