Terms of Use

1. About these terms

These Terms of Use govern two distinct things, and they are kept separate on purpose:

• The Mira software — free, open-source software released under the GNU Affero General Public License, version 3 (AGPLv3). Anyone may run, study, modify, and share it; the software and its licence are covered below.
• The hosted consultative service — the assistance offered through miramethods.org, covered below. This is a service you may optionally use; it is not the software itself.

By using the Mira software or the hosted service, you agree to the terms that apply to what you are using. If you do not agree, please do not use them. In these terms, “Mira,” “we,” and “us” refer to the maintainers of the Mira project; “you” refers to the person or institution using the software or service; and “your content” means any documents, notes, annotations, search results, drafts, and other materials you import into or create with Mira.

2. The Mira software (free and open source)

2.1 Your rights, and the licence

The software is licensed to you under the AGPLv3, whose terms control your rights to run, copy, modify, and redistribute it. Nothing in these Terms of Use limits the freedoms the AGPLv3 grants you. Where these terms and the AGPLv3 conflict as to the software, the AGPLv3 governs.

2.2 Provided “as is”

The software is provided free of charge and “as is,” without warranty of any kind, express or implied, including but not limited to warranties of merchantability, fitness for a particular purpose, accuracy, and non-infringement, as set out in the AGPLv3. Mira is research software: it can make mistakes, and its outputs must be checked against their sources. You use it at your own risk.

2.3 Your content is yours

You own your content and anything you produce with Mira. When you run the software — on your own machine or on infrastructure you control — your content and everything derived from it (stored passages, annotations, search indexes, drafts) live on your instance and remain yours. We claim no ownership of, and no rights over, your content, and we do not receive it.

2.4 Your responsibility for the rights to your material

You are responsible for ensuring you have the rights and permissions necessary for any material you import into, store in, or process with Mira. Like any research tool that stores and analyses documents — a reference manager, a qualitative-analysis package, a word processor — Mira does not acquire content on its own behalf and does not adjudicate the rights to it. In particular, you agree to comply with:

• your institution’s and your library’s agreements with publishers and other rights-holders;
• the terms of any database or subscription service you access; and
• applicable copyright law and any text-and-data-mining rules of your jurisdiction.

Mira may offer signals to help — for example, indicating when a source appears to be open access or when your selected library’s jurisdiction grants a statutory research text-and-data-mining right. These signals are good-faith decision-support, not legal advice, and they are not a warranty that any particular use is permitted. The judgement, and the responsibility, remain yours.

2.5 Acquisition tools and your own access

Mira’s optional acquisition tools (including the browser extension) operate through your own authenticated access to resources you are already entitled to reach. They make individual, human-initiated requests; they do not store your library credentials, and they are not intended for bulk or systematic downloading. You agree to use them only within your own entitlements and in keeping with the agreements referenced in section 2.4.

2.6 Acceptable use

You agree not to use Mira to: break the law; infringe others’ intellectual-property or privacy rights; circumvent access controls or authentication you are not authorised to bypass; carry out bulk or systematic downloading in breach of an agreement; or redistribute licensed content to people who are not entitled to it. You are responsible for what you do with the software.

2.7 Third-party services

Mira can connect to third-party services you choose — language-model providers, data sources, and the like. Your use of those services is governed by their own terms, and we are not responsible for them.

2.8 How Mira acquires sources, and jurisdiction

Neither a researcher nor any tool can read the hundreds of heterogeneous contracts a library holds with publishers, so Mira does not guess. It offers a signal at the moment of import, keyed to the country of the library a researcher connects, and it errs toward caution:

• Open-access, Creative Commons, and public-domain sources are acquired directly — everywhere, for everyone.
• Licensed content is acquired directly only where the researcher’s jurisdiction grants a statutory research text-and-data-mining right on lawfully-accessed works (see the table below). Everywhere else, Mira opens the article in the library’s own page, through the researcher’s normal authenticated access.
• Walled-garden platforms whose licences confine text-mining to the vendor’s own sandbox are never ingested; Mira links out to them instead.
• Where a source cannot be confirmed as clearly permitted, Mira opens the library page rather than importing. Every direct acquisition records the basis on which it was made, so the decision is auditable after the fact.

This is good-faith decision-support, not a legal determination and not a technical lock: the responsibility for the rights to any material rests with the person importing it (see section 2.4).

Whether a researcher may ingest licensed content into a local research tool is decided largely by statute, and the picture is counter-intuitive: the countries with an explicit text-and-data-mining (TDM) exception stand on firmer ground than the United States, because their statutes void any contract term to the contrary.

Publishers’ own TDM programmes come in two shapes, and Mira treats them differently. API and download programmes (Elsevier, Springer Nature, Wiley, and others) let a subscriber pull full text under a TDM licence and process it on local infrastructure — compatible with a Mira instance. Walled-garden programmes (ProQuest TDM Studio, Gale Digital Scholar Lab) require analysis inside the vendor’s environment; that content is worked in place, not ingested. An institutional arrangement is how a library resolves the grey areas explicitly, setting, for its own contracts, what may be ingested and under what limits.

Which countries fall in which group

Direct acquisition applies in the jurisdictions with a statutory research text-and-data-mining right on lawfully-accessed works:

• United Kingdom — CDPA 1988 s.29A.
• European Union and EEA — DSM Directive 2019/790, Art 3: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, and Sweden; the EEA states Iceland, Liechtenstein, and Norway; and the French region of Réunion.
• Japan — Copyright Act, Art 30-4.
• Singapore — Copyright Act 2021 (computational data analysis).
• Switzerland — Copyright Act, Art 24d.

Where no contract-overriding TDM statute exists — the United States and every other country, including Canada, Australia, New Zealand, and the rest — licensed content is not ingested automatically; it opens in the library’s own page, through the researcher’s normal access.

The same logic, already in use

This is not a novel theory. It mirrors how established research infrastructure already operates in each group: the statutory-exception countries run text-and-data-mining on lawfully-accessed content directly, while the United States and others route it through licensed or controlled environments.

• United Kingdom (s.29A): the UK government’s own copyright-exceptions guidance sets out the research TDM exception, and national services such as CORE mine scholarly texts under it.
• EU and EEA (DSM Art 3): European research infrastructures such as CLARIN and OpenAIRE run text-and-data-mining on lawfully-accessed content across member states.
• Japan (Art 30-4): the broad analysis exception underpins large-scale TDM across the research sector, including at the National Diet Library.
• Singapore (Copyright Act 2021): the computational-data-analysis exception supports national research programmes such as AI Singapore.
• Switzerland (Art 24d): the scientific-research exception is used across the institutions coordinated by swissuniversities.
• United States and elsewhere: with no contract-overriding statute, TDM runs through licensed or controlled environments — the HathiTrust Research Center, ProQuest TDM Studio, Gale Digital Scholar Lab, and JSTOR’s Constellate. Mira’s default — opening the library page rather than ingesting — matches this licensed-channel model.

3. The hosted consultative service

3.1 What it is, and what is being paid for

Mira is free and open source: anyone may run it. It is, however, technical to set up and still in development, with many rough edges. So that the methods it embodies can be developed and made accessible, miramethods.org offers a consultative service — a prefigured, ready-to-run copy of Mira, set up on a virtual machine rented on the user’s behalf (for the time being, on Google Cloud Platform), together with technical assistance in making that copy work. What is paid for is this convenience: the setup, the hosting on rented infrastructure, and the help. It is not a licence to, or a purchase of, the software — the software is free, and the same result can be had at no charge by running Mira oneself — and it is not a warranty that any particular result will be achieved. This is the ordinary way open-source work is sustained: the software is free, and a ready-to-run, hosted, or supported version is the paid service. IBM’s Red Hat sells supported, ready-to-run builds and managed environments — such as OpenShift — of open-source software; Bitnami (VMware) packages open-source applications as preconfigured, ready-to-run images; Hugging Face hosts and runs open-source models and tools on rented compute; and projects like Ghost and Discourse offer managed hosting of their own free software. In each case the fee is for the environment and the convenience, not for the software.

3.2 Who hosted access is for

Hosted access during the beta is discretionary and is not open to the general public. Even though it is being made available, it is offered selectively to active postgraduate scholars and to faculty, on a collaborative research basis — that is, to people using Mira for genuine scholarly research who are willing to work with the project as it develops. We may decline, limit, pause, or withdraw access at our discretion, and being able to pay a fee does not by itself create a right of access.

3.3 Fees

Any fees are for the ready-to-run, hosted copy and the setup assistance described above — principally to help cover the cost of the rented computing capacity the service runs on — and not a purchase of the software (which is free) or a guarantee of outcomes. Anyone who prefers can run Mira themselves at no charge. Fees, where they apply, are described at the point of sign-up.

3.4 Mira credits, and access to models

Mira credits are not required to use Mira’s features. Anyone can run the software for free, and wherever model usage is involved, a personal OpenRouter key works just as well. Mira credits are simply a convenient way to pay for things that carry a real, external cost: model usage through OpenRouter (open language models, billed per token) and additional storage capacity on the hosted service.

In future, Mira credits may also become a way to obtain access to language models fine-tuned to run in Mira’s own harness. That pathway does not exist today, and no such access is for sale: those models are still in development and do not yet exist. When they are ready, their weights will be published and made available to the scholarly community free of charge. Any fee at that point would be only for the costs of providing them — compute, electricity, and hardware, together with the associated maintenance, development, and operation — and only where we provide running versions of those models or their outputs.

3.5 Provided “as is”

The hosted service is offered in good faith and “as is,” without warranty, and may change, be interrupted, or be discontinued. It is a beta service; do not rely on it as your sole store of anything important.

3.6 Your data, and portability

Your content and configuration on the hosted service remain yours. Because Mira is open source, you are free to run it yourself and to move your work off the hosted service at any time; we will not hold your work hostage.

3.7 Suspension and termination

We may suspend or end access to the hosted service — for example, for non-payment, for use that breaks these terms or the law, or if we can no longer operate the service. Where practical, we will give reasonable notice and an opportunity to retrieve your work. You may stop using the service at any time.

3.8 What runs on Mira’s own machines, and what still relies on outside services

Being straight about where the work happens matters. Mira is built to run its whole pipeline — reading and extracting texts, and the multi-agent research itself — on its own infrastructure, and several parts already do: the passage embeddings (a self-hosted BAAI bge-m3 model producing 1024-dimensional vectors), the academic-structure extraction (GROBID), and the optical-character recognition for scanned pages (Tesseract) all run on Mira’s own servers with no third party involved.

Other steps currently rely on outside services, for one honest reason: the shared beta machine is not large enough to run everything in-house for several people at once. Specifically, the Atelier’s reasoning models (the planner, the researchers, and the synthesizer) are run through OpenRouter, billed per token; and during intake, resolving a work’s bibliographic details from its title page can use an outside vision model, alongside public scholarly catalogues (CrossRef, OpenLibrary, OpenAlex, arXiv) used to cross-check citations. These are the parts that need a large GPU or an external lookup, and running them for multiple users at once is beyond the current rented box.

This is a limitation of present capacity, not of the design. Every one of these steps has an in-house equivalent already in the code — a self-hosted vision model and self-hosted reasoning models — gated behind settings that a larger machine can switch on. As Mira secures dedicated compute (for instance through research-computing grants), these steps move onto Mira’s own hardware and off the outside services; and anyone self-hosting on a sufficiently large machine can already turn them on today. Where an outside service carries a real cost, that cost is what any fee helps to cover — it is never a charge for the free software itself.

Fair-use limits. Because capacity is shared and finite, hosted use is rate-limited — a bounded number of Atelier runs in a period, and a ceiling on how many research agents run at once — and these limits may be adjusted to keep the service usable for everyone and within its running costs. Mira credits can lift usage beyond the included allowance, and additional credits can be purchased for that purpose. The specific limits in force are shown at sign-up and in your account.

4. Privacy

Being straight about the hosted service matters more than a slogan. When you use miramethods.org, your account and your content are stored on our servers (at present, Google Cloud), and this storage is not end-to-end encrypted. That means the people who operate the service can technically access what is stored in order to run and support it. We hold that data in trust: we collect only what is needed to operate the service (such as an account identifier and basic operational logs), we do not sell it, and we do not mine or repurpose your research. If you would rather no one but you could ever reach your material, run Mira yourself — when you self-host, your content stays on your own instance and never reaches us.

Deleting your account and your data. You can delete your account at any time from your profile settings, under “Data & Privacy.” Deletion is thorough and permanent: it removes your imported PDFs and Writing-Desk documents from disk, and erases your library, its search index (the stored passages and their embeddings), your search history, your activity logs, and your account data from our database. You may either delete everything, including your email and identity, with no recovery; or erase all of your content while keeping a paid subscription tombstoned, so you can reclaim it later by signing up again with the same email.

What deletion means, and its limits. Once you delete, your material is genuinely gone as far as we are concerned: we keep no copy and retain no way to see it or bring it back. We should be just as clear about what we cannot promise. The servers this data sits on are not ours — at present they are Google’s, and we have only ever facilitated access to them, never possessed the hardware. So we can issue the deletion and lose all access to your data, but we cannot reach into the underlying storage to overwrite or physically scrub it, and we cannot rule out that recovery might be physically possible under certain forensic conditions. That final, hardware-level destruction rests with the infrastructure provider, not with us. In plain terms: after deletion it is invisible and irrecoverable to us and to anyone using Mira; what we cannot guarantee is a physical impossibility over hardware we do not control.

GDPR and CCPA. We intend to comply with applicable data-protection law, including the EU and UK General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). The account-deletion feature described above already gives effect, in the code today, to the right to erasure and the right to delete. You can also ask us to access or delete your data, or raise any privacy question, at [email protected], and a fuller privacy statement will follow.

5. Disclaimers and limitation of liability

To the fullest extent permitted by law, Mira and its maintainers are not liable for any indirect, incidental, special, consequential, or exemplary damages, or for any loss of data, arising out of your use of the software or the hosted service. The software and the service are provided “as is” and “as available.” Nothing in these terms excludes any liability that cannot lawfully be excluded.

6. Copyright complaints

If you believe content made available through the hosted service infringes your copyright, please write to [email protected] with enough detail to identify the material and your rights in it, and we will respond promptly.

7. Changes to these terms

We may update these terms from time to time. When we make a material change, we will update the date at the top of this page and, for the hosted service, take reasonable steps to let users know. Continuing to use the software or the service after a change means you accept the updated terms.

8. Governing law and contact

These terms are offered by the project’s maintainers while an entity is being formed: developers familiar with Mira are exploring incorporation as a non-profit or charitable organisation for the promotion of access to the software, and are seeking pro bono legal consultation. The United Kingdom, Germany, the Republic of Ireland, and Switzerland are presently being considered as the seat for an association that would provide the consultative services that make Mira accessible to the international research community. Once that entity is established, these terms and the hosted consultative service will be governed by the law of its seat, which this section will then name; governing law, venue, and any dispute-resolution terms remain subject to legal review. Whatever seat is chosen, two things a governing-law choice will not do. First, it will not deprive you of any mandatory consumer protections of the country where you live — where such protections apply (for example across the EU and the UK), they continue to apply regardless of the law chosen, as do applicable data-protection laws. Second, it does not change the copyright and text-and-data-mining law that governs your own use of source material: that is territorial and determined by your own jurisdiction — which is exactly what Mira’s acquisition signals reflect — and not by these terms. Questions about these terms can be sent to [email protected] or through the contact form.

9. The software licence, in plain terms

Mira is open source. The source code the software is built from is published for anyone to read. You can use Mira without paying, study exactly how it works, change it to suit you, and share it with others. None of that requires anyone’s permission.

Mira is released under the GNU Affero General Public License, version 3 — AGPLv3 for short. It is a share-alike licence: you get all the freedoms above, and in return anything built from Mira must remain open source.

10. Licence FAQ for universities, NGOs, and institutional users

If you are a compliance officer, IT lead, or grant manager evaluating Mira for your institution, this section is for you. Mira uses the AGPLv3 licence, the same licence as widely-used research software your institution very likely runs already. Institutions can use Mira with confidence, and we are happy to help with any questions or concerns about a particular use. Write to [email protected] or use the contact form and we will work through the specifics with you.

1. Why is Mira advertised as free and open source under AGPLv3, but asks beta testers to pay for access?

Mira, the software, is free and open source — it lives on GitHub, and no one charges to use it. Mira Methods (miramethods.org) provides a separate setup service: access to rented compute with Mira already installed and configured — a prefigured, ready-to-run copy — so you do not have to provision and set up the machines yourself. Relatively powerful computers are needed to run Mira and some of the large language models it depends on, and the fee covers renting and preparing that compute, not the software. You are welcome to run Mira on your own hardware for free instead; the project cannot power Mira for even one user on its current hardware. As the application improves, the hope is to acquire dedicated hardware and make the service more widely accessible.

2. Can our institution use Mira?

Yes. Researchers, faculty, staff, students, and members may use Mira freely — locally on their own machines or on infrastructure your institution operates — without owing anything to anyone. Internal use asks nothing of you: no publication, no registration, no notification.

3. Do we have to publish our research, our library, our notes, or our drafts?

No. The AGPL applies to Mira itself (the software). It does not reach your scholarship: the documents you upload, the annotations you write, the search results you produce, the drafts in your Writing Desk, and any research outputs you create with Mira are entirely yours. The license is silent on what you make with the tool.

4. Can we host Mira for our staff, students, or members?

Yes — on your own servers or any cloud you operate. If you are running an unmodified copy of Mira, nothing is asked of you beyond leaving the license notice in place, no matter who uses it or how. If the copy you host is modified, see question 6. And if you would like us to set it up for you, write to [email protected]; when we set it up, we take care of any compliance obligations that arise.

5. We charge tuition or membership fees. Does that change anything?

No. The license takes no interest in whether your users pay you. A university whose students pay tuition, or a library whose members pay dues, may run Mira for them, and may run a modified Mira for them, on the same terms as anyone else. The only thing the license asks, and only when the copy you run is a modified one, is that the modified version be shared: share it with us and we make it available to the whole Mira community. Nothing more.

6. What if we modify Mira?

You can modify Mira however you like. The license simply requires that the modified version be shared: share it with us, and we make it available to the whole Mira community, so everyone has the opportunity to benefit from any upgrades. Write to [email protected] or open a pull request and we fold your improvements into the main Mira repository on GitHub.

7. What if we host a modified Mira for others to use?

You can. The only thing asked is the same sharing described in question 6: share the modified version with us and we make it available to the whole Mira community.

8. Do we have to open-source our other systems, or anything else we own?

No.

9. What is an example of a use of Mira that is not allowed under the license?

Suppose an IT team modifies Mira into a product of its own, hides the source code of the modified version, and sells access to it as a new subscription service. That is the use the license does not allow.

10. Can you help us integrate Mira with our existing systems?

Yes. We are glad to work alongside your IT team to connect Mira with the systems you already run: catalogues, repositories, sign-on, and the rest. We will manage any and all licensing and compliance issues that come with the work, so your team never has to become licensing experts. Write to [email protected] or use the contact form to start the conversation.

11. Can someone take Mira, reskin it, and sell it as their own closed product?

No.

12. We have a specific concern. Who can we talk to?

Write to [email protected] or use the contact form and describe the use case. You can also open an issue on the project’s repository at github.com/MarxReader. We are happy to work with universities, libraries, and NGOs to confirm that a particular deployment plan is uncomplicated.

11. Why AGPLv3?

Mira is a research tool for scholars and the institutions that support them. AGPLv3 keeps the project genuinely open: any improvements that end up in a publicly-hosted version come back to the community, so the tool can’t be quietly captured and offered back as a closed service. For the people Mira is built for — researchers, libraries, archives, and the NGOs that work alongside them — this is exactly the protection they need.

12. Full text of the licence