.png)
micro1
micro1 is a San Francisco data lab supplying frontier AI labs with expert human data, reinforcement-learning environments and contextual agent evaluations across three offerings — Realm, Cortex and Robotics — sold entirely through direct commercial contact rather than self-service plans.
What is micro1?
micro1, Inc. is a data lab based at 655 Montgomery Street in San Francisco. It describes itself as building the infrastructure for advancing intelligence through expert human data, real-world training environments and contextual evaluations, and it sells to frontier AI labs, Fortune 10 enterprises and US government agencies rather than to individual users.
The work is organised under three product lines. Realm is the training and evaluation engine: reinforcement-learning environments that mirror real-world scenarios, plus expert data across more than a hundred research domains including finance, medicine, law, coding, STEM, vision-language models and audio. Three internal components support it — Zara, an AI recruiter agent that sources and vets domain experts; Merit, a dashboard quantifying pipeline quality, velocity and reliability; and Flow, the environment where experts create, review and deliver datasets. Clients get live access to model performance and pipeline health, with tracking of velocity, error rates, cost per task and quality.
Cortex addresses agents already in production. It runs a four-step loop: designing success criteria and scoring rubrics, diagnosing failures through domain-expert review that goes beyond automated evals and LLM-as-judge, producing targeted training data for the highest-priority failure modes, then re-evaluating as the system evolves.
Robotics covers expert-demonstrated capture in point-of-view and third-person formats, dataset structuring with labelling, tracking, 2D and 3D annotation and QC, and a full-stack deployment lab where robots are run in controlled environments. Capture is multimodal: synchronised stereo video, IMU and multi-camera.
Two adjacent programmes complete the picture. Data partnerships invite companies to turn anonymised operational workflows into recurring revenue, with announced tiers of 100,000, 500,000 and more than a million US dollars a year. A government offering carries Awardable status through the CDAO's Tradewinds Solutions Marketplace.
Behind all of it sits a network of more than 130,000 vetted candidates spanning over a hundred domains and sixty languages. micro1 has published reasoning benchmarks for legal, pathology-report and financial tasks, raised a 35 million dollar Series A in September 2025 at a 500 million dollar valuation, and claims a run rate above 200 million dollars in annual recurring revenue.
What it does
- Produce expert human training data across more than 100 specialist domains
- Build reinforcement-learning environments that mirror real-world agentic scenarios
- Design evaluation criteria, rubrics and scoring for a given agent task
- Diagnose where, why, when and how a production agent fails, through domain-expert review
- Monitor agent reliability continuously as models, prompts, workflows and rules change
- Capture and annotate real-world robotics data for vision-language-action training
- Recruit, vet and certify domain experts through an AI interviewer
When to use micro1 / When not to
A quick filter to help you decide if micro1 is the right fit.
When to use micro1
- Frontier AI labs that need expert-generated training data, reinforcement-learning environments and published reasoning benchmarks
- Product and engineering teams shipping agents into production who must measure, diagnose and monitor failure modes
- Robotics labs sourcing expert-demonstrated capture, 2D and 3D annotation and controlled deployment environments
- Operationally mature companies of roughly 30 employees or more looking to monetise anonymised internal workflows
- Domain experts — lawyers, physicians, financial analysts, linguists, engineers, cinematographers — seeking paid, fully remote AI training work
When not to use micro1
- Individuals and small teams wanting a self-service tool: there is no plan, no public price and no online sign-up for clients
- Developers expecting an API or an SDK, since micro1 publishes no developer documentation of any kind
- Companies below roughly 30 employees, or without documented internal processes, which fall outside the data partnership eligibility criteria
- Regulated buyers who need compliance artefacts up front: no DPA, no subprocessor list and no security certification is published
- Teams operating outside English, as data partnerships require English documentation and English-speaking representatives
How to use micro1
A typical end-to-end flow, from setup to results.
- Decide which side you are approaching from: buying data services, contributing company data, or applying as a domain expert
- For client work, open the 'Get in touch' form and give your contact details, a description of the project and how you found micro1
- Expect a scoping conversation: there is no self-service path, no price list and no online checkout
- For a data partnership, start with discovery and share workflow details so micro1 can assess their potential value
- Move to partnership design: agree approved workflows, privacy boundaries, anonymisation, contribution formats and internal approval processes
- Begin data distribution, contributing approved workflow examples on a recurring basis while the company operates normally
- For Cortex, start with evaluation design and define success criteria, rubrics and scoring for the agent's task
- Let domain experts diagnose failures, then commission targeted training data against the priority failure modes
- Re-evaluate and monitor performance as models, prompts, workflows and rules change
- To join as an expert instead, apply to an opportunity, take the 20 to 40 minute Zara AI interview, get certified, onboard, and follow your work on talent.micro1.ai
Pros & Cons
Pros
- A very large vetted expert network: more than 130,000 candidates across over 100 domains and 60 languages
- Three published reasoning benchmarks — legal, pathology-report and financial — make part of the quality claim independently checkable
- Named case studies, including Box, rather than anonymous logos
- Substantial commercial traction: a 35 million dollar Series A at a 500 million dollar valuation and a claimed run rate above 200 million dollars ARR
- Awardable status on the CDAO Tradewinds Solutions Marketplace simplifies procurement for US federal agencies
- Data partnership confidentiality commitments are unusually specific: negotiated scope, anonymisation, isolated pipelines, sample review and deletion
- Contributing companies keep ownership of their underlying data, and LLM vendors are barred from training on candidate data
Cons
- No public pricing, no plans and no trial: every engagement starts with a sales form
- No public API and no developer documentation of any kind
- No Data Processing Agreement is published or offered, and no subprocessor list is disclosed
- No security certification is displayed — neither SOC 2, nor ISO 27001, nor HIPAA
- The word GDPR appears nowhere on the site, even though EU and UK data rights are granted
- No terms of service cover the website or the client services; only candidate and referral terms exist
- Retention is described only as 'no longer than necessary', with no figure published anywhere
Pricing & Plans
micro1 publishes no prices. There is no pricing page, no plan structure, no permanent free plan and no free trial: every client engagement — Realm, Cortex, Robotics or the government offering — begins with the 'Get in touch' form and is quoted individually. The monetary figures that do appear on the site run the other way, being amounts micro1 pays out. Data partnerships are presented with potential tiers of 100,000, 500,000 and more than 1,000,000 US dollars in annual recurring revenue for the contributing company, determined case by case according to dataset size, workflow complexity, data quality, domain expertise and the rarity of the operations involved. Experts are paid hourly or per task, with the amount shown on each posting. Referrers receive up to 3,000 US dollars per successful placement and up to 50,000 US dollars for a company referral.
- Realm — frontier evaluations and reinforcement-learning environments for AI labs
- quoted on request
- Cortex — agent evaluation
- failure diagnosis and post-launch monitoring
- quoted on request
- Robotics — expert-demonstrated capture
- annotation and deployment-lab work
- quoted on request
- Data partnerships — a programme in which micro1 pays the contributing company
- scoped and priced individually
- Government — agent buildouts for federal workflows
- procurable through CDAO Tradewinds Awardable status
- application
- AI interview and certification carry no fee
Data, GDPR & hosting
A consolidated view of how micro1 handles your data.
GDPR overview
The acronym GDPR appears nowhere on micro1's site — a search across the text and archived HTML of nineteen pages returns no occurrence — and the company never claims compliance in so many words. What it does publish is narrower but concrete. Residents of the EU and the UK are granted access, a machine-readable copy, confirmation of processing, rectification, erasure, portability, restriction and objection, and withdrawal of consent, and may lodge a complaint with their local supervisory authority. Four legal bases are stated: contract, legitimate interests, legal obligation and consent. For international transfers micro1 says it may sign model clauses. A Data Protection Officer is reachable at dpo@micro1.ai and rights requests go to privacy@micro1.ai. No Article 27 EU representative is designated. Both privacy policies are dated 21 July 2026.
Who owns the data?
Under the data partnership programme, contributing companies retain ownership of their underlying data. micro1 structures approved workflow examples and delivers them to frontier AI labs, but the scope is agreed before anything moves: the company defines which workflows are approved, what must be redacted or anonymised, which security standards apply and who signs off internally. Companies may review representative samples before prepared data is used, and deletion is honoured on request or at the end of an engagement. Candidates keep control of their own profile and can delete their information at any time, subject to the legal retention obligations micro1 reserves.
Reuse rights
Onward use of data is micro1's business, so reuse runs in both directions and is governed by contract rather than by any public licence. Approved, anonymised workflow examples are structured by micro1 and supplied to frontier AI labs for model training; personally identifiable information is removed where relevant, access is restricted to a limited number of authorised staff, engagements run through isolated pipelines so client data is not commingled, and some datasets undergo synthetic rewrites that preserve the workflow structure while loosening the tie to original records. For candidates the position is narrower and stated explicitly: micro1 reserves the right to use de-identified interview material — audio, video, transcripts and assessment results — to train and enhance its own proprietary machine learning models, but says it does not share candidate data, de-identified or otherwise, with AI model companies for the training of their models, and that the large language model providers it uses process candidate information solely on micro1's behalf and never to train their own models. Nothing published on the site grants an end user a standing right to reuse data without permission.
Data retention & training
Hosting summary
micro1 states that it is located in the United States and that information provided to it is transmitted to and processed in the United States, as well as in other locations where micro1, its vendors or its third-party partners operate. Protection therefore falls under United States law, which the privacy policy acknowledges may differ from the law of a user's own country. For international transfers the company says it may sign model clauses in accordance with European Union and other data protection laws. Cloud hosting and data storage appear in the list of vendor categories, but no provider is named, no region is committed to and no data-residency option is offered. The public site itself resolves to 198.202.211.1, an anycast Cloudflare address routed through a San Francisco node, which describes the delivery layer rather than where client datasets live. In short: United States jurisdiction, no contractual guarantee of hosting location, and no published detail about the infrastructure holding customer or candidate data.
Where micro1 works
Country-level availability.
Not available in
Things to keep in mind
Risks and trade-offs to weigh before adopting micro1.
- Nothing about cost, scope or contract terms can be known before a sales conversation, which makes budgeting and comparison difficult
- Ownership is retained on paper, but the point of a data partnership is that your operational data leaves the organisation and trains someone else's model: the scope is only as tight as the contract you negotiate
- Candidates should understand that de-identified interview audio, video, transcripts and assessment results may be used to train micro1's own models
- With no DPA, no subprocessor list and no security certification, a regulated buyer has no published artefact to rely on and must obtain every guarantee contractually
- Liability is capped at 100 US dollars in the published candidate and referral terms, with disputes going to JAMS arbitration under Delaware law
- Expert income is project-dependent: certification confers eligibility, not placement, and payouts depend on role availability and client demand
- Referral earnings can encourage volume over fit; attribution works only through the referral link and pays once, on the first successful match
Setup & Integrations
Technical difficulty
There is nothing to install. micro1 delivers a service rather than software: no API, no SDK and no integration work. The effort is contractual and organisational instead of technical. A data partnership requires a negotiated design phase covering approved workflows, anonymisation rules, contribution formats and internal sign-off. A Cortex engagement requires an evaluation design phase before anything can be measured, since success criteria, rubrics and scoring must be defined first. For an expert, the entry cost is a 20 to 40 minute AI interview, occasionally followed by a short skills assessment.
Deployment
Supported languages
Behind micro1
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
How much does micro1 cost?
Who does micro1 sell to?
What are Realm, Cortex and Robotics?
How do I join micro1 as a domain expert?
Does micro1 recruit experts everywhere?
Who owns data contributed through a data partnership?
Does micro1 train AI models on customer data?
How much does the referral programme pay?
Does micro1 offer an API?
Does micro1 publish a DPA or a subprocessor list?
Should you pick micro1?
micro1 occupies a clearly defined position in the human-data market: it recruits, vets and pays domain experts, then turns their judgement into training data, reinforcement-learning environments and agent evaluations for frontier AI labs, large enterprises and US government agencies. The three product lines — Realm, Cortex and Robotics — are genuinely distinct rather than repackaged, and the company backs its claims with published reasoning benchmarks in law, pathology reporting and finance, named case studies including Box, a 35 million dollar Series A at a 500 million dollar valuation, and Awardable status on the CDAO Tradewinds Solutions Marketplace.
Product transparency is therefore reasonable. Contractual transparency is not. There is no pricing of any kind, no API, no Data Processing Agreement, no subprocessor list and no displayed security certification; the word GDPR does not appear anywhere on the site even though EU and UK data rights are granted; retention is described only as 'no longer than necessary', with no figure; and the only terms micro1 publishes cover candidates and referrers, not clients. Liability in those documents is capped at 100 US dollars.
None of this makes micro1 a poor supplier. Enterprise data work is normally negotiated contract by contract, and the confidentiality commitments described for data partnerships are unusually specific. It does mean nothing can be assessed before a sales conversation, which rules the company out for anyone hoping to evaluate it self-service.
The symmetry is worth noting. micro1 is one of the few AI companies whose offer is genuinely two-sided: enterprises buy expertise, and lawyers, physicians, analysts, linguists, engineers and cinematographers are paid for supplying it. For a qualified domain expert, that side of the platform is open, free to join and immediately concrete.
- Choosing a selection results in a full page refresh.
- Opens in a new window.