Jobbit Projects: Give Your AI Agent Shared Files, Instructions, Secrets and a Team

The first thing you learn about working with an AI agent is that it is only as good as what it knows about you. The second is how tiring it is to tell it again. Your price list, your brand voice, the contract template, the API key for the newsletter tool, the fact that the client prefers British spelling: every new chat started from zero. Jobbit Projects end that. A project gives a group of chats the same files, instructions, notes and secrets, so every conversation starts with the knowledge already in place, and a team can share it.

This guide explains what Jobbit Projects are, how project knowledge works (including what "searchable inside" and the token allowance mean), how instructions, secrets, notes, members and join links fit together, how to fork a chat to continue it, and how Projects compare with ChatGPT Projects, Claude Projects, Google NotebookLM, Notion AI and Gemini Gems. New to agents? Start with what is agentic AI?.

What a project is

In Jobbit, a project is a container that holds:

• Files. Documents, spreadsheets, PDFs, images, exports: the material the agent should work from.

• Instructions. Standing guidance in plain English: who you are, how you like things done, what the agent must never do.

• Secrets. API keys and credentials the agent may use for automations and builds, stored for the project and never pasted into chats.

• Notes. Short written memory with a byline: decisions, context, things the team should know, rewritable at any time.

• Chats. Every chat started in the project works with everything the project holds, and you can search, pin and fork them.

• Members. People invited by join link, with shared projects showing up under "Shared with you".

Think of it as the difference between briefing a freelancer from scratch every morning and giving a colleague a desk with the folder already on it.

Project knowledge: files the agent can actually use

Drop files into a project and Jobbit reads them so that any chat in the project can search inside them. Each file shows its state:

• Findable by name. Just added; the agent knows it exists.

• Waiting to be read through. Indexing is in progress.

• Searchable inside. The agent can find and quote specific passages, tables and figures.

• Partly searchable. The file was too long to read in full; the first part is searchable and the agent tells you so.

You can also add a short description to any file ("Revenue by product line for the third quarter") so the agent picks the right document faster, and search across the files you have picked. Under the hood this is retrieval-augmented generation, explained in the AI agent glossary, packaged so you never have to think about embeddings.

The allowance, explained in pages

Project knowledge is measured in tokens, the units models read in, and the project page shows the room left as a ring rather than a bar. A token is about three quarters of an English word and a page of prose is roughly five hundred words, so a project allowance of five million tokens is on the order of several thousand pages, and a single document is capped at a tenth of that so one giant file cannot crowd out the rest. The exact ceilings come from your plan; the page says how much of the project a file will take before you add it. Plan details are in Jobbit pricing explained.

Instructions: your standing brief

Instructions are the standing brief every chat in the project inherits. Good ones read like the six-part brief from how to write prompts for AI agents, minus the one-off task:

• Who you are. "A two-person accounting firm in Manchester serving tradespeople."

• Voice and rules. "British English, plain words, no jargon, never promise a refund."

• Defaults. "Quotes in our template, prices from the attached rate card, all dates in day-month-year."

• Boundaries. "Draft emails but never send them. Ask before spending anything."

Write them once and every chat, automation and build in the project starts from them. Change them and the change applies everywhere.

Secrets: keys the agent can use without seeing them in chat

Automations and apps need credentials: a payment key, an email API token, a database password. Pasting them into a chat is a bad habit. Project secrets store them by name, paged and searchable, so the agent can use "the Stripe key" when it builds your checkout or "the Mailgun token" when it sets up the weekly report, and the value never appears in a conversation. Secrets belong to the project, so a build started in the project has what it needs from the first message. The security reasoning is covered in vibe coding mistakes and security, where "no secrets in the code" is rule three.

Notes: memory you can read

Notes are short written entries with the author's name on them: the client's preferences, a decision taken in a call, the reason the budget moved. The agent reads them alongside the files, and people read them too, which makes a project a shared memory for the team rather than a private history inside one person's chat. Notes can be rewritten as things change, and the project counter reflects them.

Members and join links: working with other people

Projects are built for teams. The owner creates a join link, optionally set to expire after a while or after a number of uses, and sends it to colleagues or a client. People who join see the project under "Shared with you", work with the same files, instructions and notes, and start chats inside it. Links can be turned off or replaced at any time, and expired or used-up links say so plainly rather than failing silently. Leaving a project is one click.

For an agency this means one project per client, with the client's files and voice loaded once and every team member working from the same brief. For a founder it means a project per product, with the investor deck, the roadmap and the customer feedback where every chat can reach them.

Chats in a project: search, pin and fork

Every chat started in a project works with everything the project holds. The project page lists its chats with who started them, lets you search them, pin the ones that matter and page through the rest. Two things are worth knowing:

• Fork to continue. Any chat can be forked, from a message or as a whole, into a new chat that keeps the history and notes where it branched. It is how you explore two directions from the same starting point, or hand a colleague a copy to take further. Forking is explained in steer your AI agent mid-run.

• Ask from anywhere. You can ask the project a question from the page you are on, and the answer draws on the whole project.

What to put in your first project

A practical starter set for a small business or freelancer:

1. Instructions: who you are, your voice, your defaults and your boundaries, in ten lines.

2. Files: price list, service descriptions, two example proposals, brand guidelines, the current contract template.

3. Notes: the three things new team members always get wrong.

4. Secrets: the keys for the tools you want automated (payments, email, calendar).

5. A first chat: "Draft a quote for the job described below, in our template, from our rate card." Compare it with what you would have written.

From there, the automations in 50 AI automation ideas for small businesses become one-line requests inside the project, because the context is already loaded.

How Jobbit Projects compare

The difference is what the chats can do once they know your context. In Jobbit a project-aware chat can build and deploy the client portal, run the invoice chase every Friday, generate the product photos and write the newsletter, using the same files, instructions and secrets. That is the agentic half of the story, covered in the future of AI agents.

Frequently asked questions

What is a Jobbit Project?

A project is a container that gives a group of chats the same files, instructions, notes and secrets, with members invited by join link. Every chat started inside it works with everything the project holds, so the agent starts each conversation already knowing your business.

How much can I store in a project?

Project knowledge is measured in tokens; a page of prose is about five hundred words, and the project page shows how much room is left and how much a file will take before you add it. Allowances depend on your plan, and one document may not take more than a tenth of a project on its own.

Can other people work in my project?

Yes. Create a join link, optionally with an expiry or a limited number of uses, and share it. Members see the project under "Shared with you", use the same files, instructions and notes, and start their own chats inside it. Links can be turned off or replaced at any time.

What are project secrets?

Secrets are credentials such as API keys stored by name in the project so the agent can use them in builds and automations without the values ever appearing in a chat. They are paged and searchable, and they belong to the project rather than to one conversation.

How do Jobbit Projects differ from ChatGPT Projects or Claude Projects?

All three let you attach files and instructions to a set of chats. Jobbit adds secrets for automations, join links for teams, and chats that can build and deploy software, run automations and generate media from the shared context, rather than only answering in the conversation.

Load the context once and let every chat start from it. Start free on Jobbit and create your first project.