Blankline

Blankline is an AI research and product company whose stated mission is to build safe AI for humanity's multi-planetary future. According to its own materials, it was founded in 2024 by a group of researchers from leading AI labs who shared the conviction that the most powerful AI systems should also be the safest. The company positions safety not as a brake on capability but as a constraint that shapes how capable systems are built, describing its aim as making truly capable AI interpretable, verifiable, and aligned with human values. Blankline operates on two connected fronts. The first is applied research: it runs research across four domains — AI Safety, Astrophysics, Biology, and Physics — with small teams that publish openly. Its AI Safety work covers alignment, evaluation, and runtime safety for agentic AI systems, and the company says this work underlies every Dropstone tier. Its Astrophysics work includes fast radio burst analysis and wormhole signature detection pipelines run end-to-end on internal AI agents; Biology covers protein structure modeling and computational biology workflows powered by tool-augmented reasoning; and Physics covers computational complexity, statistical mechanics, and theoretical physics under an AI-first methodology. The company also states it works to lower the environmental cost of that research. The second front is its commercial product line, Dropstone. Dropstone is an agentic coding CLI that runs frontier open-weight models behind an approval gate on every tool call. It is offered in three tiers — Dropstone Fast, Dropstone Pro, and Dropstone Heavy — sharing one runtime, one approval gate, US-hosted inference, and a zero-retention guarantee. Fast is a low-latency model for inline completion and short edits; Pro is the default coding model; and Heavy is a long-horizon model for repository-scale work and agent swarms. The tiers are refreshed monthly as the open-weight leaderboard moves, and the company frames the choice as picking the smallest model that meets the task. Blankline also publishes the Joule Index, described as an honest per-dollar comparison of coding agents that normalizes every plan to the same coding-turn definition so price-per-capability is directly comparable across providers. Its public positioning combines a safety-and-transparency ethos — publishing failures, open-sourcing methods, and describing itself as a 'glass box' — with a developer-tools business built on open-weight models hosted in the US.

Blankline was founded in 2024 by a group of researchers from leading AI labs. According to the company's own account, they came together around a shared conviction that the most important AI systems would also need to be the most trustworthy, and they saw an industry racing toward capability while treating safety as an afterthought. Blankline was founded to prove a different path was possible: that truly capable AI systems must be interpretable, verifiable, and aligned with human values — not in spite of their capabilities, but because of them. The company frames its mission as building safe AI for humanity's multi-planetary future and its vision as securing the digital foundation of the modern world. As AI becomes integrated into critical infrastructure, Blankline says it is building the verifiable logic layers that ensure these systems remain reliable, robust, and aligned with human intent. Its stated values include treating safety as a non-negotiable constraint (when safety and capability conflict, the model is paused), radical transparency (publishing failures and open-sourcing methods), an intergenerational view that rejects decisions that hurt the long term, and scientific integrity requiring every claim to be verifiable. On the research side, Blankline reports running research across four domains — AI Safety, Astrophysics, Biology, and Physics — with 21 active research areas and small teams that publish openly. Its AI Safety work covers alignment, evaluation, and runtime safety for agentic AI systems and is described as underlying every Dropstone tier. Its Astrophysics pipelines include fast radio burst analysis and wormhole signature detection run end-to-end on internal AI agents. Its Biology work covers protein structure modeling and computational biology workflows powered by tool-augmented reasoning, and its Physics work covers computational complexity, statistical mechanics, and theoretical physics under an AI-first methodology. The company also states it applies AI to lowering the environmental cost of research. Commercially, Blankline ships Dropstone, an agentic coding CLI that runs frontier open-weight models behind an approval gate on every tool call, with US-hosted inference and a zero-retention guarantee. Dropstone is offered in three tiers — Fast, Pro, and Heavy — sharing one runtime and refreshed monthly as the open-weight leaderboard moves. The company describes Fast as a low-latency tier for inline completion and short edits, Pro as the default coding model, and Heavy as a long-horizon model for repository-scale work and agent swarms. It also publishes the Joule Index, a per-dollar comparison of coding agents normalized to a common coding-turn definition. The company's public materials emphasize that it is a technology company building safe AI, and its product and research pages present an integrated picture of a safety-focused lab that also sells developer tooling.

• 2024

  Blankline is founded by a group of researchers from leading AI labs.

• Blankline launches Dropstone, an approval-gated agentic coding CLI running frontier open-weight models.

• Blankline publishes the Joule Index, a per-dollar comparison of coding agents.

• Dropstone 1.7 introduces three tiers (Fast, Pro, Heavy) with monthly refreshes; last cycle's Heavy tier becomes the default Pro tier on every plan.

Safety-first AI research company and developer-tools provider; approval-gated, US-hosted, open-weight coding agents with radical transparency.

• Dropstone

  Agentic coding CLI that runs frontier open-weight models behind an approval gate on every tool call, with US-hosted inference and zero retention.

• Dropstone Fast

  Low-latency tier for inline completion and short edits; built on DeepSeek V4 Flash 0731, hosted in the US.

• Dropstone Pro

  Default coding model tier; built on GLM-5.2, hosted in the US.

• Dropstone Heavy

  Long-horizon tier for repository-scale work and agent swarms; built on Kimi K3, hosted in the US.

• Joule Index

  A per-dollar comparison of coding agents that normalizes every plan to the same coding-turn definition.

1. 1Blanklineblankline.org
2. 2About Us — Building Safe AI for Humanity | Blanklineblankline.org
3. 3Dropstone Models: Fast, Pro, Heavy | Blanklineblankline.org
4. 4Blankline — Wikidatawikidata.org