NanoGPT Projects Are Becoming AI Workspaces

Attaching a file to one chat is easy. Keeping a larger piece of work organized across many chats is harder.

NanoGPT Projects help by keeping the important parts of a project together: files, instructions, notes, tasks, and previous conversations. A new chat can pick up the same body of work without making you upload everything again.

Recent improvements to file reading and editing make Projects feel less like folders of attachments and more like working spaces for research, writing, planning, and smaller software projects.

What a Project keeps together

A Project can include:

• Files such as documents, code, specifications, and drafts
• Instructions that should apply across the Project
• Notes for decisions and useful facts
• Tasks for open and completed work
• Conversations that can be searched or revisited later

This means one conversation can focus on research, another on drafting, and another on review while all three use the same Project context.

There is also a Project isolation option. When it is enabled, NanoGPT tells the model to focus on the current chat and this Project's files and conversations instead of using global memory or outside web information unless you provide it.

Models can read project files more reliably

File analysis can go wrong when the model receives only a shortened or reformatted version of a file.

NanoGPT now makes a clearer distinction between two ways of reading:

• A quick document view helps the model find relevant information.
• An exact text view reads the original text or code when the stored file is available and can be decoded safely.

The exact view is useful when you want the model to check a configuration value, quote a passage, inspect code, or suggest an edit based on the real source text.

Long files and documents can be read in sections instead of stopping after the first excerpt. Very large files and extracted documents still use the document view rather than pretending the text is an exact copy of the original.

Clear warnings when the source is not exact

Sometimes NanoGPT cannot retrieve the original file and has to use extracted or reconstructed text. A file may also contain characters that cannot be decoded cleanly.

That fallback can still be useful, but formatting, spacing, or individual characters may differ from the source.

Project file cards include a Download action. The editor and download flow also warn when the content is reconstructed or contains replacement characters, so you can check the source before trusting a quote or saving an edit.

This matters most for code, configuration files, and other material where one character can change the meaning.

Edit files without rebuilding the Project

Text and code files can be edited directly in the browser.

You can also upload a replacement version of a non-media file while keeping the same Project file. Existing Project references continue to point to it instead of forcing you to add a separate item and reorganize the workspace.

Audio and video files are not replaced in place; upload those as new files.

Generated content can move back into the Project too:

• Save a code block from a model response as a Project file.
• Let a capable model propose a new or replacement text file.

Model-proposed files always require confirmation. NanoGPT shows the filename and content, warns if an existing file will be replaced, and lets you save, replace, or decline the change.

Previous conversations remain useful

A capable model can search and read saved conversations from the same Project.

That is useful when an earlier chat contains a decision, an unfinished plan, or research that the current conversation needs. The model can search by topic, list available conversations, open a specific one, or retrieve the most recent conversation when you ask to continue earlier work.

NanoGPT also asks the model to include a short Referenced section when it uses Project files or conversations. When the model follows that instruction, you can see which sources supported the answer.

Three practical uses

Software work

Keep a small set of source files, requirements, configuration, and implementation notes together. Ask a model to explain a bug or suggest a patch, then review the proposed file before saving it.

Projects are not source control. They do not provide branches, merges, version history, or an automatic test runner. Keep Git as the source of truth for a real repository.

Research and analysis

Upload papers, reports, transcripts, and notes. Keep important findings in Project notes and unanswered questions in tasks. A later conversation can return to an earlier source or compare it with a previous discussion.

Check important quotes against the original document, especially when the source came from OCR or extracted PDF text.

Writing and planning

Store the brief, outline, terminology, drafts, and editorial rules in one Project. Put lasting requirements in Project instructions, short-term decisions in notes, and unfinished sections in tasks.

Separate conversations can handle research, structure, editing, and final review without losing the shared context.

Choose a model that can use Project tools

The full Project experience works best with models that can call tools reliably. Those tools let the model search files, read documents, open attachments, retrieve previous conversations, and propose files.

A model without good tool support may only see the Project information already placed directly in its prompt. For large documents or code-heavy work, reliable tool use can matter as much as the model's general writing or coding ability.

Use a vision-capable model when asking what is visible in an image. Otherwise the model may only have the filename or surrounding text to work from.

What Projects are not

Projects organize persistent context, but they are not autonomous development environments.

Models do not receive unrestricted access to your filesystem. File writes are limited and require confirmation. Extracted documents may not be exact. Models still have context limits, and web searches or other tools may send data through additional services.

Projects sit between a one-off chat and a fully autonomous agent: they keep work organized across conversations while leaving important file changes under your control.

Open Projects, create a Project, and add the files and instructions you want available in later conversations.