> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hired.tools/llms.txt
> Use this file to discover all available pages before exploring further.
# The brain
> Everything you know about your own career, stored raw and searched rather than read. Roles, brain dumps, highlights, notes, standing rules and the four supporting collections.
The brain is the part of Hired that makes the rest work. A resume builder with nothing
behind it is a text editor; a resume builder sitting on ten thousand words of what you
actually did is something else.
Nothing in here is a document. It is raw material, and it is searched rather than read.
## Roles
A role is one job you have had. It holds:
| Field | Notes |
| ---------------------- | ----------------------------------------------------------- |
| `company`, `title` | The only two required to create one. |
| `employmentType` | Full-time, Contract, Internship, Freelance — a free string. |
| `location` | City, state, or `Remote`. |
| `startDate`, `endDate` | `YYYY-MM`. Leave `endDate` empty for a current job. |
| `isCurrent` | Whether this is the job you have now. |
| `summary` | One or two sentences on the scope of the role. |
| `brainDump` | The point of the whole thing. Unlimited, markdown welcome. |
| `tags` | Freeform, e.g. `['fintech','ic','python']`. |
### The brain dump
Unlimited free-form text about that job. Projects, metrics, technologies, stories,
praise, org context, the thing that went badly. No structure required, no length limit,
and deliberately no formatting rules.
Length is a feature. This is the raw material every future resume is assembled from, and
the difference between a thin one and a thorough one is the difference between a resume
that names the number and one that says "responsible for".
Write it the way you would tell it to someone:
> Owned billing infra. The nightly reconciliation job took six hours and paged the
> on-call every Tuesday — rewrote the pipeline in Go and it came down to twenty minutes.
> Team of four, I was the only one who had touched the old Python version. Also did the
> Stripe migration in Q3, which nobody wanted, and it shipped a week early.
**Add, do not overwrite.** `append_role_brain_dump` puts new material on the end, with
an optional markdown heading to file it under. `update_role` with a `brainDump`
replaces the entire thing. The append tool exists specifically because the update one
was eating people's notes.
## Highlights
A highlight is one polished, reusable achievement bullet — the distilled version of
something in a brain dump, written in resume voice: strong verb, specific scope,
quantified outcome.
Each carries `text`, an optional `impact` if the number is not already inside the text,
`tags` for retrieval, and `strength` from 1 to 5. They come back strongest first, and
they are what a resume gets assembled from when there is no time to mine the raw dump
again.
You do not have to write them by hand. The [`mine_brain_dump`](/workflows#mine-a-brain-dump-into-highlights)
workflow reads a role's raw dump, checks what highlights already exist so it does not
duplicate them, and writes the rest in one call — flagging anywhere the dump hints at
impact without giving a number, so you can fill it in.
A highlight can be archived rather than deleted, which keeps it out of the way without
losing it.
## Notes and standing rules
Notes are free-floating material that belongs to no single job: STAR stories, interview
prep, references, compensation history, anything. They can be pinned and tagged, and they
are found by searching.
A note has a `kind`, and this is where it gets interesting.
### Standing rules
Set `kind: "GUARDRAIL"` and the note stops being something to find and becomes something
that is always true. Guardrails are **carried in the briefing every connected assistant
receives**, at the end of it, so they are in context before a single tool is called.
Use them for the things that must never be got wrong:
* How you may and may not be described. *"I was a contributor to the migration, not its
lead. Do not describe me as having led it."*
* Numbers that are unsettled and must not be cited. *"The 40% figure was never verified.
Never put it on a document."*
* Credit that must not be overstated. *"Do not claim the patent — I am one of five
names on it."*
Everything else is a `NOTE`, which is the default.
Guardrails are the mechanism for the failure that actually happens. Nobody's assistant
fabricates a job from nothing. It upgrades a credit into a hire because the posting
asked for one. A guardrail is how you make that specific upgrade impossible.
## The supporting collections
Four collections handled by the same four tools — `list_extras`, `create_extra`,
`update_extra`, `delete_extra` — with `kind` selecting which one:
`school`, `degree`, `field`, `gpa`, `details`, `location`, `startDate`, `endDate`.
`name`, `role`, `url`, `description`, `brainDump`, `startDate`, `endDate`, `tags`.
Projects get their own long-form brain dump, exactly like roles do.
`name` is the group name; `skills` is the list inside it, e.g.
`['Python','Go','Rust']`. Passing `skills` to `update_extra` replaces the whole list.
`name`, `issuer`, `date`, `url`.
## The profile
One identity block per account: name, headline, email, phone, location, website,
LinkedIn, GitHub, X, a career summary used as the default resume summary — and a personal
brain dump of its own, for the things that are about you rather than about a job. Values,
what you want next, compensation expectations, non-negotiables.
### The photo
One picture serves the whole app. It is your avatar in the interface *and* the headshot
on every resume whose design has the photo switched on, so changing it once updates every
document at once and there is never a second copy to keep in sync.
Set it in **Settings → Account** — drop a file in and drag it around the circle — or ask
for it: *"use the photo on my GitHub profile."* An assistant can pass an `https` URL for
the server to fetch, or the bytes inline as a data URI. Anything over 400KB is refused,
which is far more than a resume needs: it prints about an inch square.
Whether a *given* resume shows it is a separate decision — `showPhoto` on that document,
or **Design → Photo** in the editor. It is off by default, and the Harvard template never
renders one whatever you set.
`get_profile` and every tool that returns a profile report `hasPhoto` rather than the
picture. The bytes are hundreds of kilobytes of base64, and flooding an assistant's
context with an image it cannot look at helps nobody.
## Getting things out again
Two tools, and the choice between them matters:
Ranked keyword search across brain dumps, highlights, notes, projects and the
profile. Returns excerpts with the id and kind of each hit, so you can fetch the
full record. **This is the first call when tailoring anything.**
Everything at once: profile, every role with its full dump, every highlight,
education, projects, skills, certifications and notes. Right for writing a resume
from scratch; large enough that targeted lookups should use search instead.
Snapshot takes `include_brain_dumps: false` for a lighter payload when you only need the
shape of a career rather than its contents.
## In the app
**Me** lists every role with its date range and highlight count. Opening one gives you
the brain dump in a full-width editor that autosaves, its highlights beside it, and the
role's own fields. The supporting collections live on the same screen.
Everything here is also reachable by conversation — see [the brain tools](/tools/brain).