Ordos builds autonomous AI agents that orchestrate complex scientific workflows end-to-end, so researchers can focus on discovery, not data wrangling.
"The tools already exist. The orchestration layer between a scientist and those tools does not."
Ordos closes that gap. We build specialized agents, each connected to the best scientific models for its domain, designed to be used by scientists directly, not just engineers.
Each agent connects a specific scientific domain to the models, workflows, and context it actually needs.
Scientists describe research tasks in plain language. Osmo builds the workflow, presents it for approval, and executes using AlphaFold2, Chai-1, Boltz-2, ESM C, and Evo 2. Results include interactive 3D structures and ranked compound lists.
Autonomous materials discovery, synthesis planning, and property prediction.
Autonomous genomic analysis, variant effect scoring, and sequence design.
Every workflow plan is shown to the scientist before anything executes. You review, modify, approve. Nothing runs without your explicit sign-off.
Local PII filtering, OS keychain key storage, tamper-evident audit logs.
Generic AI fails at precision science. Each agent is built around the exact models, context, and workflows its domain requires, nothing repurposed, nothing generic.
Join researchers already on the waitlist. We onboard every team personally.
Type your research objective in plain English. Osmo understands scientific intent, no structured inputs, no query syntax.
Osmo presents a complete step-by-step workflow, which models it will call, in what order, what outputs to expect. You see everything before execution.
Nothing executes without your sign-off. Modify the plan if needed. Then approve, and Osmo takes it from there.
Interactive 3D protein structures, ranked compound lists, binding affinity scores, and exportable structure files. Delivered to your desktop and optionally to Slack or Discord.
Osmo connects to domain-leading scientific models via the BioLM API. No compromises on scientific quality.
Osmo selects the right model for each step automatically. Multiple models, zero configuration.
Your most sensitive data stays on your machine.
Interactive protein visualization with per-residue pLDDT confidence color-scoring. Export as PDB or CIF format.
Every workflow plan is presented for review before execution. Modify it, reject it, or approve it.
Every action logged. Reproducible, exportable, and verifiable, built for compliance.
Desktop app for macOS, Windows, and Linux. No browser required. Results optionally delivered to Slack or Discord.
One platform. Multiple agents. Each purpose-built for its domain, connected to the right models, designed for scientists who don't want to write code.
Osmo is a desktop AI agent for biotech research labs and drug discovery teams. A scientist describes what they want to do in plain English, for example, "Predict the 3D structure of this KRAS G12C mutant and screen known inhibitors by binding affinity." Osmo parses the task, builds a full step-by-step workflow, presents it for approval, then executes autonomously.
Results include interactive 3D protein structure visualizations with per-residue pLDDT confidence scoring, binding affinity scores and binding probability for screened compounds, ranked compound lists, and a summary report. Scientists can export structure files in PDB or CIF format and review a complete audit trail of every action taken.
Designed for the entire lab, bench scientists, medicinal chemists, PIs, and biotech founders, not just computational biologists. No coding background required.
Orsin is being built for materials scientists and engineers. It will enable autonomous workflows for materials discovery, property prediction, and synthesis planning — connecting to specialized models for computational materials research.
Like Osmo, Orsin will require no engineering resources to run complex research workflows. Materials scientists will describe what they need in plain language, approve the plan, and receive structured results.
Odin is being built for genomics researchers and synthetic biologists. It will enable autonomous workflows for genomic sequence analysis, variant effect scoring, gene design, and synthetic biology applications.
Odin will connect to specialized genomics models and allow biologists to run complex sequence analysis and design workflows directly — without writing a single line of code.
It deserves tools that don't get in the way.
A single drug discovery experiment requires dozens of disconnected tools, engineering resources, and hours of manual data wrangling. Brilliant scientists spend more time managing pipelines than doing science.
The specialized AI models exist, AlphaFold2, Chai-1, Boltz-2, Evo 2. They are extraordinary at what they do. But they have no orchestration layer. Researchers still need to know which model to use, how to format inputs, how to interpret outputs, and how to chain results between tools. That is a full-time engineering job on top of the actual science.
The result: the most powerful computational biology tools in the world remain locked behind an engineering barrier that most bench scientists cannot cross.
Generic AI tools fail in science because science requires precision. The same model that drafts emails cannot reliably predict protein-ligand binding affinities — and pretending otherwise produces results no scientist should trust.
Each Ordos agent is purpose-built for its domain. It knows the right models for each task, the right order to call them, how to handle failures, and how to synthesize results into something actionable. It is not a chatbot. It is an autonomous collaborator designed for a specific scientific discipline.
The human approval layer is not a safety feature bolted on after the fact. It is the design. Scientists make decisions. Agents do the execution. That division of labor is what makes Ordos useful in a real lab.
Scientists see every step before it runs. No black boxes. The agent explains its complete workflow plan and waits for approval before touching a single model API.
Proprietary research data never passes through Ordos. Local PII filtering, direct API connections, OS keychain key storage, and tamper-evident audit logs give scientists full data sovereignty.
The agent does the execution. The scientist makes the decisions. Ordos is optimized for researcher leverage — not researcher replacement. Every design choice reflects that.
One agent at a time.
"The scientists doing the most important work in the world should not be blocked by engineering overhead. We built Ordos to remove that barrier permanently — not by replacing scientists, but by giving them agents that handle everything that isn't the insight itself."
