Dropstone

Dropstone is a self-hosted AI assistant built by Blankline, an AI research lab founded in 2024 and headquartered in Chennai, India. It combines a coding agent, a chat interface, a CLI, and real-world action capabilities — all backed by persistent long-term memory that spans every surface. The platform runs on open-weight models hosted by Blankline or deployable on your own infrastructure.

What It Is

Dropstone is a multi-surface AI agent designed to be proactive rather than reactive. Unlike conventional AI assistants that wait for a prompt and forget context between sessions, Dropstone maintains a single persistent memory across the CLI, chat, SDK, and phone calls. It started as a terminal-based coding agent and has since expanded to monitor inboxes, control smart home devices, place phone calls, and ship code — all with approval gates on every tool call.

Architecture and Deployment Model

Dropstone is built for self-hosting. According to the product page, the entire stack — keys, machines, and network — can run inside your own infrastructure with nothing leaving your walls. The platform is model-agnostic by design, supporting open-weight models, self-hosted models, and local models through Ollama. Blankline publishes that it runs Kimi K3 behind the "Dropstone Heavy" tier, which it claims is the highest-scoring open-weight model on the Artificial Analysis Index v4.1.1. A one-million-token context window is available on every tier.

Surfaces and Capabilities

Dropstone spans multiple interaction surfaces:

• CLI — terminal-native coding agent that writes and ships code
• Chat — web-based interface with file and photo uploads, voice mode, and long-term memory
• SDK — developer integration layer
• Real-World AI — phone calls placed on your behalf, inbox monitoring, and smart home device control
• Figma Plugin — the open-source Dropstone Design Bridge connects Figma to the agent for AI-assisted design creation and editing using 36 real-time design tools

The Figma plugin (released as V1.0.1 in March 2026) lets users select any Figma element, describe a change in plain language, and have the AI apply targeted edits with a visual feedback loop — the agent screenshots the result and self-corrects until the design looks right.

Self-Learning and Research Backing

Blankline describes a self-learning behavior in Dropstone where the agent reflects on completed work, identifies patterns, and applies them to future tasks without being explicitly retaught. The company states it ran this as a formal study and observed the same habit generalize to mathematical pattern recognition. Research findings are published openly at blankline.org/research.

Update: v3.0.6 and Recent Release History

According to the About page timeline, Dropstone shipped its initial public release in August 2025 and has moved through several major versions:

• v2.0 (September 2025) — Semantic analysis engine, 40+ language support
• v2.0.3 (November 2025) — D2 Engine, advanced memory system
• v3.0.1 (December 2025) — D3 Engine runtime, major architecture upgrade
• v3.0.4 (January 2026) — Stability fixes, security patches
• v3.0.5 (February 2026) — Share Chat, Team Memory, collaborative workspaces
• v3.0.6 (February 2026) — Thought Chain, latency optimizations, model updates

The Figma plugin (Dropstone Design Bridge V1.0.1) was published on March 9, 2026, under an MIT license on GitHub.

Privacy and Security Posture

Blankline publishes a safety-first charter: zero prompt, code, or output retention; no data shared with the model's original operator; no routing through labs that train on customer data. The infrastructure is operated and audited by Blankline directly, with personal audit logs available to users. The company states that model training on user data is "strictly no" across all tiers.