Auxot
Tutorials Docs Pricing Login

Tutorial 04

Add your first context file

Paste a chunk of your business — brand voice, sales playbook, meeting minutes — into a context file, attach it to your agent, and watch the next reply change shape.

Plus: three power moves that turn a context file from "the agent has my doc" into "the agent has institutional memory I never had time to use myself."

Audience Everyone · Executives
Time ~5 min
Prerequisites An Auxot account on any tier. At least one piece of business writing handy (a doc, a few pasted Slack threads, meeting minutes — anything you'd hand a new hire on day one).
You'll end up with One context file attached to your agent — and a clear sense of why this is the single most leverage-multiplying thing you can do for an agent.

Why this matters

In Tutorial 01, we said your agents become immensely more powerful the moment they have context files. This is that moment.

Think of a context file as the briefing you’d write for a new hire if you ever had the time — the rules, the patterns, the institutional shortcuts, the do’s and don’ts that took your team years to figure out. Anything you’d hand someone on day one to make them effective. That’s what goes into a context file.

You hand the agent that briefing, and from that point on, when you ask it a question, you’re not getting generic AI — you’re getting your business. The questions don’t change. The answers do. Without the file, the agent’s guessing at what your company might want. With it, the agent’s the new hire who walked in day one with your manual already memorized — and is answering like someone who’s been on your team for a year.

One context file teaches an agent more about your business than ten Slack threads explaining things. The leverage compounds as you add more.

Today, add your first context file. Tomorrow, the same agent that gave you a stock answer yesterday is answering from a business — yours.

Quick start

  1. Sign in — open Auxot in your browser and log in.
  2. Open Context Files — click Context Files in the left menu.
  3. Click New File — top right of the page.
  4. Fill in the form — give the file a title (something like “Our brand voice” or “Q1 sales notes”), paste the content into the Content box, and pick which agent(s) should use it. Save.
  5. Test it — open chat with the agent and watch the file shape the next reply. (If you pasted brand voice rules, ask the agent to draft an email in your brand voice.)

Done? Your agent now has a piece of your business attached. Every relevant conversation from here pulls from it automatically.

The agent can do that?

A context file is just a doc until you ask the agent to do something interesting with it. These three prompts each show a different kind of leverage you can pull from the same file.

1. The meeting-minutes drift detector

Take your last three to six months of meeting minutes — one context file per meeting type (weekly leadership minutes in one file, all-hands in another, departmental reviews in a third — single-topic files give you sharper answers and keep token costs down). Attach all of them to an agent, and ask:

Based on these meeting minutes, what topics or initiatives have we stopped talking about? What did we commit to that nobody's checked back on? What's drifted from where we said we'd be three months ago?

Then, for the three most important things you found, give me a concrete next step for each — who should own it, what the smallest forward action is, and what would unblock it. Make each recommendation specific enough that I can act on it in this week's meeting.

Why it’s non-obvious: Drift on its own is just an uncomfortable list. The leverage is in what comes next — who picks it back up, what the smallest forward step looks like, what’s been blocking it. The agent reads every word of your minutes in thirty seconds, drafts the who/what/blockers for each drift item, and hands the whole thing back to you. You decide what to act on. The agent does the legwork; you make the call.

2. The voice that knows your business

Take your brand voice rules, your customer support FAQs, and a few sample emails your team has actually written — three separate context files (one per topic), all attached to the same agent. Then ask:

Draft a reply to this customer email [paste the customer's email], using our brand voice and the patterns from the example emails I've given you. Don't invent details — only use what's in the context files.

Why it’s non-obvious: Generic AI drafts sound like generic AI. The same draft from an agent with your voice files reads like your team wrote it — same tone, same hedging, same sign-offs. Stop rewriting AI drafts to sound like you. Hand the agent the files and let it sound like you from the start.

3. The pre-flight reviewer

The inverted-usage move. Take a draft you’ve already written — a contract clause, an internal memo, a customer email, the talking points for a slide deck — and give it to the agent alongside your existing context files (compliance rules, brand voice, anti-patterns from past reviews). Ask:

Compare this draft against the context files I've given you. What does the draft say or imply that contradicts the rules in the context files? What would our most careful reviewer flag first?

Why it’s non-obvious: You become your own toughest reviewer, on demand, with full institutional memory backing you up. You’re not asking generic AI for generic feedback — you’re asking an agent with your specific rules to find your specific conflicts. The draft gets better in two minutes.

Go deeper

How to get content into a context file (paste, not upload)

Despite the name, you don’t upload a file from your computer. You paste the text content directly into the Content box on the New File form. Behind the scenes, that pasted text becomes the “file” your agent uses.

The workflow: open your source material — a Word doc, a PDF, a Google Doc, a web page, a notes app — copy the text, paste it into the Content box. For most cases this is actually faster than uploading, because you can grab just the part that matters and skip the rest.

File upload may arrive in a future version, but it isn’t required to do any of this today.

Inline vs Indexed — when to pick which

The page’s own helper text describes them this way:

Indexed: “Outline is shown in the prompt. The agent searches the full content with tools when it needs specifics. Best for large files.”

Inline: “Full content is pasted directly into every prompt. Best for small, critical snippets the agent must always have in view.”

Practical translation:

  • Inline is for things you want the agent to always be thinking about, no matter what you’re asking. Brand voice rules. Hard constraints (“never recommend X”). The principles you want present in every single reply.

  • Indexed is for one-topic reference material the agent should consult when relevant but not pre-load every time. The full spec for one product. The complete playbook for one process. A detailed FAQ for one customer segment. The agent searches the full content when the conversation calls for it, instead of hauling the whole thing into every reply. (Resist the urge to lump multiple topics into one Indexed file just because Indexed handles long files — single-topic-per-file is still the recommended way to get the sharpest answers.)

The actual question isn’t “how long is this file?” — it’s “does the agent need this in front of it every conversation, or only when the topic comes up?” Always-needed content goes Inline (even if it’s longer than feels efficient). Look-it-up-when-relevant content goes Indexed.

Troubleshooting — the agent isn’t using my file
  • The file isn’t attached to the agent you’re chatting with. Open the file in Context Files, check the agents list, make sure the agent you’re testing with is included. (Or open the agent’s settings and confirm the file is attached there.)
  • The agent has the file but isn’t reaching for it. This usually means the question isn’t specific enough to make the agent search. Ask something more pointed, or reference the file directly — “Using our brand voice rules, draft this email…” tends to work better than “Draft this email.”
  • You picked Indexed and the file is small. Indexed mode searches the file when needed; for small files, the agent may not “need” to search and might answer from general knowledge instead. Switch the file to Inline.
  • The file content is structured oddly. The agent reads what you pasted as plain text. Bullet lists, headings, and clear sections help it find the right part. A wall of unstructured text gives it less to grab onto.
Variations & edge cases
  • One file or many? Splitting content into smaller, well-named files generally outperforms one giant file. “Brand voice” and “Email examples” as two separate files is better than “Marketing reference” with both inside.
  • Editing files later is fine and live — change the content, save, and the agent picks up the new version on its next reply. (You don’t need to re-attach to agents.)
  • Personal vs. Org scope is locked at creation. If you create a Personal file and later realize the whole team should see it, the way to fix that is to create a new Org-scoped file with the same content and delete the Personal one.
  • Org files require admin permissions on multi-user accounts. On Free tier, you’re the admin, so you can create either kind. On Team/Business/Enterprise, non-admin users can only create Personal files — for shared knowledge they want available org-wide, they’d ask whoever admins the account.
  • Files attached to multiple agents all stay in sync. Update once, every agent that uses it gets the new version.
  • Letting the Admin Agent create the file for you. You can ask the Admin Agent to draft a context file from a description and create it on the spot — useful when you’d rather talk through what should go in than write it yourself. One thing to be ready for: it saves the file the moment you give it the go-ahead, so have your description ready before you ask. Read it back afterward; treat it like a draft you’d review from a teammate.
  • Critiquing existing files through chat. If you ask the Admin Agent to review a context file you already have, it can see the title and a content preview — not the full contents. For a real critique, paste the file’s text back into chat alongside your question. The agent will read every word of what you paste; it just can’t pull the full body of an existing file on its own.

Walkthrough

Step 1: Sign in

Open Auxot in your browser and sign in.

Step 2: Open Context Files

Click Context Files in the left menu. The page’s own subtitle frames it well: “Background knowledge for your agents — like a handbook or style guide they should follow.” That’s exactly the right way to think about a context file.

You’ll see two tabs near the top: Org Files (knowledge shared with your whole team) and Personal Files (knowledge just for your own conversations). For your first one, Personal Files is the simplest start — anyone can create one regardless of role or tier.

Step 3: Click New File

Click New File in the top right. A form opens.

Step 4: Fill in the form

A few fields, none of them complicated. Here’s what each one means in plain English:

a. Title. Whatever you’ll recognize this file by later. “Our brand voice,” “Q1 sales numbers,” “Customer support FAQs,” “September meeting minutes” — descriptive is better than clever. (You can always rename later.)

b. Content. Paste the text of your document here. Yes, paste — there’s no file upload in this version. If your content is in a Word doc, a PDF, or a Google Doc, copy it out and paste it in. Markdown formatting works but isn’t required. There’s no character limit you’d practically hit.

Tip: One topic per context file is the strong recommendation — it’s how you get the most out of context files and keep your conversations efficient. Even if your source covers a lot (say, a 200-page employee handbook spanning hiring, expenses, PTO, and remote work), you’ll get sharper, faster answers if you split it into separate context files by topic — “Hiring policies,” “Expense rules,” “PTO.” Single-topic files give the agent better focus AND keep every conversation faster and cheaper. The same logic applies to meeting minutes (one file per recurring meeting type tends to work better than all of them lumped together) and customer interviews (one file per theme rather than the whole archive).

c. Scope. Two choices:

  • Personal — just for your own conversations. Other people on your team don’t see your file or its contents. This is the right choice for things only you need (your personal style, your specific notes, drafts you’re working on).
  • Org — shared across your whole team. Useful for things the whole team should access — your house style, escalation rules, the company brand voice. Creating Org files requires admin permissions on your account; if you’re on Free tier, you’re the admin, so you can create either kind.

For your first file, Personal is the simplest start. You can always create more files of either kind later. (Note: scope is locked in at creation — you can’t switch a file from Personal to Org after the fact. You’d just create a new one.)

d. Agents. Pick which agents this file applies to. You can attach a file to one agent or several. Don’t have any custom agents yet? Leave this blank for now — you’ll come back and attach the file once you’ve built your first custom agent (Tutorial 07).

e. Injection mode. This sounds technical but the choice is simple:

  • Inline — the full content of the file is included in every prompt to the agent. Best for the things the agent must NEVER forget: your brand voice, hard rules (“never offer discounts”), how the agent should and shouldn’t respond. Inline content is in front of the agent every single conversation, no exceptions.
  • Indexed — the file’s outline goes into every prompt; the agent looks up the full content when a question calls for it. Best for one-topic reference material that’s detailed enough to bloat every conversation if it were Inline — say, the full spec for one product, the complete playbook for one onboarding process, or all the test results from a single experiment. (One topic per file is the recommended way to get the most out of context files — see the Tip below.)

If you’re not sure, ask yourself: does the agent need this in front of it every single conversation, or only when the topic comes up? Every time = Inline. On demand = Indexed. You can change this later.

Step 5: Save and test it

Click Save. Your file appears in the Context Files list under the scope you picked.

Now — the moment of truth. Open chat with the agent you attached the file to and ask it something the file should shape. For example:

  • If you pasted brand voice rules: “Draft an email to a customer who’s been waiting too long for an order. Use our brand voice.”
  • If you pasted Q1 sales notes: “What was Q1’s biggest revenue driver?”
  • If you pasted last week’s meeting minutes: “What did we agree to do in last Thursday’s meeting?”

If the answer is shaped by what you pasted, you’re connected. From here on, every related conversation with this agent pulls from your file automatically — no extra clicks, no reminders.

If the answer still reads generic, see Go deeper → Troubleshooting below. Usually it’s either the file attached to a different agent than the one you’re chatting with, or a question that wasn’t pointed enough to make the agent reach for it.

What’s next

Reference

  • Pages in Auxot: Context Files, Settings → Agents
  • Two scopes: Personal (just for you) and Org (whole team — admin-only to create)
  • Two injection modes: Inline (always in the prompt) and Indexed (searched when relevant)
  • Content entry: paste only (no file upload yet); Markdown supported
  • See also: Tutorial 05: Give your agent its job description, Tutorial 07: Create an agent from scratch