Student · Source-check notes
Organise assignment sources and your own notes in a project. Check missing citations with the matcher first, then record which sources need a closer look as tasks in Studio.
Run a preliminary citation checkYour manuscript, tasks and AI team in one Windows workspace. Break the review into tasks, follow the work and review the result.
Setup, models and your first task
Free app · Windows 10/11 · 64-bit 1.1.4
Questions about setup or use? Browse the FAQ
AI use requires eligible model access. Your provider’s terms, quotas and any API charges apply.
A quick look at Studio. Sound starts off.
A short Studio introduction showing agents, documents and tasks in one workspace. AI features require an eligible model connection, and users must verify the outputs.
Microsoft Store
Open Studio as a standalone Windows app.
Store version 1.1.4 is available. Install it from Microsoft Store or update your existing app from the same link. The Windows app is distributed through Microsoft Store.
If you previously used the standalone edition, export your work before moving to the Store edition. Back up before uninstalling or resetting the app.
Studio can use your computer’s official Codex or Claude Code session. Sign in with an eligible subscription first. One model can serve the team, or you can assign a different model to each agent.
Access and usage limits depend on your provider. There is no automatic fallback to a paid API.
Studio · Task workflow
Add your document, specify which sections to review and start the task. The work record shows agent stages together with any incomplete checks.
See the first stepsHow do I start?
Install the Windows app from Microsoft Store, then open Studio. Install from the Store
Create a project and add the sources you need. Before running an agent, check the model connection supported by your edition and your provider’s usage limit. Adding a file alone does not start a model request.
Select the relevant document and a clear instruction. Sending the task shares the selected content with your provider. Review the output and export your work before switching editions or uninstalling.
Organise assignment sources and your own notes in a project. Check missing citations with the matcher first, then record which sources need a closer look as tasks in Studio.
Run a preliminary citation checkKeep verified teaching sources and synthetic examples together. Ask an agent for question ideas, then review the suggestions and their supporting evidence yourself. Do not share student identifiers.
Compare DOI metadataSplit method, reference and reporting checks into separate tasks. Treat agent suggestions as review notes, not decisions; verify them against original sources. Obtain the necessary permission before sharing confidential documents.
Open the submission checklistLooking for an answer?
Choose a topic or search for the control, file type or problem you are working with. Answers refer to Studio 1.1.1 and link to the relevant guide section.
Enter one or more words, or choose a topic below.
Showing 42 of 42 questions.
Use the AcademicFix Studio listing in Microsoft Store. This guide covers the Windows desktop app. The previous web workspace and standalone EXE/ZIP distribution are not the current public installation route.
The package is x64 and declares Windows 10 build 19041 or later. It uses a native window with Microsoft Edge WebView2 and .NET Framework 4.7.2. Separate features can need additional components: a local model server, an analysis runtime/packages, or Word/configured LibreOffice for the DOCX page-rendering workflow.
This guide does not establish that every Windows device configuration has been tested.
No application account is needed to open and organise the local interface. AI work needs your own eligible provider access or a configured local model. Studio does not include a model subscription or API credit.
One tested model can serve the team. Agents using one account share its limits; separate agent requests still consume that account’s usage.
Use a small nonconfidential sample that you are authorised to process. Test and assign a model, import the sample, inspect its extracted text, then ask for a narrow review. Read the proposed outputs before approving production and inspect the downloaded files afterward.
Check the version card in Settings. This guide describes Studio 1.1.1. The chosen language, theme, motion setting, window size and project history can also affect what is visible. Some controls appear only when a relevant record or permission exists.
They are separate account routes in Models. ChatGPT uses official account sign-in with eligible plan access. Local Codex checks an official Codex installation already signed in with ChatGPT. Claude Code checks the official local tool and its account, with a consented setup/login flow where needed.
After sign-in, select and test an accessible model, then assign it to the team. A successful account sign-in alone is not a successful model response.
Access depends on the provider, account eligibility, permissions, available model and current limits. Studio’s connection screen and real response test establish what that connection can do at that time.
A plan name or a previous successful test is not a guarantee for a future request. If access is denied, inspect the specific connection and the provider’s account information.
A direct consumer-subscription connection for Gemini or Antigravity is not available. A Gemini API key or the experimental Google OAuth route uses separate Gemini API/project access and quota. It does not use your chat subscription and may incur API charges.
The experimental OAuth option needs a Google Cloud desktop-app test setup; it is not the normal substitute for a chat subscription.
No automatic paid-API fallback is used. An optional API connection is a deliberate separate choice and can have separate billing.
The provider may also have extra credits or overage settings on your account. Review those settings yourself if you do not want additional provider charges; the absence of API fallback does not turn off provider overage.
Yes, the interface offers Ollama and compatible local-server routes, including an LM Studio preset. Start the server, load an appropriate model, check the local base address and run a short response test in Studio.
Local inference depends on your computer and model. It does not automatically supply web access, every document capability or an image-capable reviewer, and it does not automatically disable requested web tools.
It is a recorded successful response from that model at the displayed time. It is historical evidence, distinct from a test performed in the present session.
If the account, key, model or usage conditions changed, refresh the model list and test again. Listing a model, signing in and assigning it are separate from obtaining a real response.
Test the desired model, then open the team assignment step. The default preserves existing individual overrides. Inspect which agents will change and which will be preserved.
Use the explicit replace-all option only if you want to replace every assignment. Changing a connection’s default does not mean every previously customised agent has been reassigned.
The Usage view displays provider-reported account limits. Several agents can share the same account. Message count or token usage is not a reliable measure of remaining subscription allowance.
If the provider does not report a value, or the last measurement is stale, Studio leaves it unknown. Refresh Usage and check the provider’s own account page; the app does not assume an expired period means 100% has reset.
Develop ideas supports discussion and permitted reading without creating or revising files or running analysis. Production prepares concrete work, but still requires you to approve the exact proposed outputs.
The choice is saved for the project. An already running job retains its recorded approval scope.
The question records what work and files you authorise. Read the output list and source scope, then choose Approve and produce or Continue discussing.
A mode selection, model statement or instruction inside a source document cannot replace your decision. Changed source material or team instructions may require a new proposal.
Saving a task creates the project record. Start it from Tasks when the source, assigned model and required permissions are ready. A production or missing-information question may need your answer first.
Open the existing work detail before creating a replacement task, so you do not duplicate an already queued or paused request.
Use the Chief Editor when the task needs coordinated planning, several specialists, separate reviews or a combined report. Use a direct specialist conversation for a narrow question within that role.
Common chat goes to the coordinator; it is not an automatic broadcast to every agent.
A direct message addresses one agent. A topic room keeps a topic, 2–6 active participants, shared-document selection and history together; each message can address the group or one recipient. Room configuration supports up to eight documents.
A meeting is a focused two-agent discussion with one or two rounds and its own recorded conversation. These views organise different forms of work; they do not create new provider allowances.
Open the related work and inspect completed stages, questions, connection errors and existing files. Repair the relevant connection or supply the requested information, then use continuation controls where available.
Uncertain work is not automatically replayed after an interruption. Retry manually only after checking the record, especially if a request may already have produced a file.
No. They are configured AI roles making separate model calls. Different agents can use the same underlying model. Review and criticism records help you inspect the process, but do not establish human peer review or independent external-source verification.
Use the actual evidence, accessed sources and output checks to assess the result.
The picker includes text and structured-text formats, DOCX/PDF, XLSX/XLS/ODS, selected statistical-data formats, research code/notebooks and ZIP. The import section lists the supported families.
Each file may be up to 50 MiB and extracted text is limited to 500,000 characters. Read import notes: accepted format does not mean every feature, object or nested archive file was fully processed.
Automatic OCR for scanned PDFs is not included. A PDF with useful text can be imported for reading/review; an image-only scan may need an authorised OCR-processed or text version.
The preserved-original revision pipeline requires an imported original DOCX and creates a new working copy. PDF import is not a direct-edit PDF revision feature.
Automatic document access can use the shared board documents, or the room’s shared selection in a topic room. The composer shows the documents for the current message.
Choose selected-only access to narrow the scope, or clear the selection for no documents. Excluding a chip changes the current message; Remove from project archives the document record. Adding a file by itself does not start a model request.
The implemented original-DOCX revision flow preserves the imported source and produces a new working copy. You can download the original from Documents and inspect revised files in Outputs.
State what should change and what must remain unchanged. Local analysis likewise uses copies of the selected inputs, but generated analysis code still runs with your Windows account’s privileges.
Use Requested output files for the intended delivery and Supporting files for intermediate material, work records and access notes. Read the version and review labels. Latest created can differ from latest AI-reviewed.
Preview content or actual PDF pages where available, then download the integrity-verified file. A missing or changed file is not offered as a verified download. Open Related work if a promised format or review is missing.
No. Despite that button label, the JSON export contains saved workspace state across projects, not only the active project. It excludes separate credentials but can include confidential work content from other projects.
It is not a complete portable archive of original document binaries, and a one-click JSON import/restore is not described by this version. Inspect the export before sharing it and make a private application-data backup before reset/uninstall.
No. Analysis checks the available Python, R or JavaScript runtime and uses the installed environment and required packages. Inspect runtime/version information and the task’s saved logs.
The bundled document parser/export runtime supports its document functions; it does not prove that every statistical environment or package is present.
It lets agents in the active project execute generated local analysis code. The setting is off by default and applies to the project team.
Code runs with your Windows account’s filesystem and network permissions; project organisation is not an operating-system sandbox. Enable it only for projects you trust. Desktop window access is a separate opt-in.
The graphical pilot is limited to Rgui 4.6.1 with separate local-execution and desktop permissions. RStudio graphical automation is not verified; this version has no SPSS/JASP/jamovi GUI integration.
Importing a data format does not establish installation, licensing or desktop control of the program associated with it. Command-based analysis is separate.
No. Successful execution establishes that the program completed under the recorded conditions. File checks establish expected output availability and recorded integrity.
Numerical comparison can check selected saved values from two successful runs with different scripts and matching inputs under stated tolerances. It does not establish assumptions, suitable model choice, causal interpretation or scientific validity.
No. Storage and the interface are local, while requested cloud-model and web operations can send selected context to the enabled services. Local-model use does not automatically disable web operations.
Review the provider, document scope and processing permission before starting work with confidential or unpublished material.
By default the API key stays in memory while the app is open. Remembering it is optional and uses Windows protection for the current Windows user. Use the connection’s Forget key control to remove the remembered key.
Separate stored keys and account session credentials are excluded from the work-record JSON export. That exclusion does not remove confidential research content from the export.
Remove in Documents or Tasks archives the record and offers restoration. Clear project lets you select documents, conversations and tasks separately; cleanup history can undo a cleanup.
Original files, generated outputs, agents and model connections are retained by this cleanup. It is reversible workspace organisation, not secure erasure. Finish or stop active work before removing records.
Close Studio and privately back up the application data and files you need. App data can include confidential documents and protected sessions; store the backup securely.
A JSON records export is useful evidence but is not a complete document-binary package or promised one-click recovery mechanism. Do not assume reset or uninstall will preserve everything you need.
The Store edition uses a separate profile from earlier standalone EXE/ZIP or web profiles. Automatic synchronisation between them or across devices is not provided by this workflow.
Check which installation contains the actual work and preserve it before changing or removing an installation.
Decorative idle movements and coffee reactions do not call a model, change instructions or delay the work queue. A file handoff animation follows a real recorded delegation event; it is not evidence on its own that the delegated task finished.
Real quality meetings and agent discussions make model requests through their work flows. Read their records for participants, progress and outcomes.
Use the top Music button to turn music on or off. For volume, open Sound or Settings → Music and sound, then use the Music slider and mute control.
Five local tracks play in order and music lowers during speech. Audio choices are retained on reopening. If nothing plays, check channel settings, the computer’s output and whether an app interaction is needed to unlock audio.
Yes. In version 1.1.1, music and sound effects have independent sliders and mute buttons. Muting effects also controls the welcome effect, without muting music or meeting speech.
The main Music button affects music only. Use the separate speech playback controls for meeting voices.
Speech uses voices available on the Windows computer. Open Voice options, refresh the list and preview the chosen voice. If the needed language is unavailable, add a suitable Windows speech voice and refresh.
The written meeting record remains available without speech. A voice from another language may pronounce the text poorly; selecting it does not translate the conversation.
No. The described voice feature speaks actual agent text using local voices. It does not open the microphone or send text to an external speech service.
Model requests remain separate and follow the selected provider connection. Stop audio controls playback; stopping the model meeting requires the work-stop control.
No. It changes the interface and supported default material. Your messages, documents, custom names and custom instructions remain as written.
Ask for the desired output language in the work request. Choose an appropriate installed speech voice separately if you want to listen.
It evaluates completed outputs against a saved rubric and can prepare reasoned instruction changes from groups of five completed evaluations. Inspect evidence, before/after text and rollback options rather than treating an average score as an objective accuracy measure.
Automatic evaluation is optional, off by default, and works while the app service is running. It changes saved instructions, not provider model weights, and does not continue as a cloud job after closing Studio.
1.1.4 · Microsoft Store
Version 1.1.4 is available in Microsoft Store. Update your existing app through the Store, and continue to inspect output files and any incomplete checks in the work record.
AI use requires a supported provider account and eligible access. There is no automatic fallback to a paid API. Gemini chat subscription connections are not supported.
Help shape Studio
Report a bug, suggest a feature or share your experience. No account needed; follow progress with your private code.
Work records stay on your device. Uploading a file does not start an AI request; sending a task or message shares selected documents with the chosen provider. You can export your records from the app.
Explore security and architecture