Overview

Your account at a glance — who you are, what you've used, and where you're signed in.

Profile

How you appear in Q, and which model new topics should start with.

Your details

Your username is your sign-in identity and appears in shared links and administrative records, so it can't be changed here.

Username

Display name

Default model

Used for new topics. "Let Q choose" follows the deployment default, so you keep up with changes without doing anything.

Security & devices

Your password, the protection on your sign-in, and every device this account is currently signed in on.

Change password

Changing your password here does not sign out your other devices — use "Sign out everywhere else" below if that's what you want.

Current password

New password

Repeat new password

Signed-in devices

Session identifiers are never shown — the list uses a reference that can't be used to sign in.

Device	Address	Signed in	Last seen	Expires

API keys

Credentials that let an editor, a script or another service talk to Q as you — without your password.

Where keys come from

Using a key in a coding editor

Continue, Roo Code, Cline, Zed and anything else with an "OpenAI-compatible" or "custom endpoint" option all take the same two values:

Base URL: https://q.planqtec.com/v1
Model: q-v2
API key: one of yours below, sent as a normal Authorization: Bearer header — every OpenAI-compatible client does this without being told to.

For Continue specifically, this goes in ~/.continue/config.yaml:

models:
  - name: Q V2
    provider: openai
    model: q-v2
    apiBase: https://q.planqtec.com/v1
    apiKey: YOUR_KEY_HERE
    roles: [chat, edit, apply]
    capabilities: [image_input, tool_use]

Using Continue over VS Code's Remote-SSH (connecting to another machine over SSH and wanting Q to run commands there and read back the result)? Continue itself needs one small, one-time fix first — without it, commands run but their output never comes back. Run whichever matches your OS once after installing Continue, and again after any Continue update:

Windows:

powershell -ExecutionPolicy
                        Bypass -File continue-remote-ssh-patch.ps1

. Mac/Linux:

perl
                        continue-remote-ssh-patch.pl

— Perl ships by default on both, nothing extra to install. Neither script touches your Q account or your key; both only edit Continue's own installed files, back up the original first, and change nothing if they don't find exactly what they expect.

Your keys

The key itself is never shown here. Q Cloud stores only a hash of it, so a lost key cannot be recovered — it has to be revoked and reissued. Revoking takes effect on the next request the key makes.

Key	Can do	State	Created	Last used	Expires

What the scopes mean

A key holds only the scopes it was issued with. None of them reach the admin area, and none of them reach anybody else's data.

My topics

Every conversation you own. Deleting one here removes its messages from the server too, not just the row.

Open Q

	Title	Workspace	Model	Created	Updated

My files

What you sent to Q and what Q made for you. Only you can download these.

Files you sent

"Extraction" is whether Q could read text out of the file. A failure is reported with its reason rather than hidden.

	File	Type	Size	Extraction	Sent

Files Q made for you

	File	Format	Created

Account activity

Administrative actions taken on your account — approvals, password resets, status changes.

Which administrator acted is deliberately not shown here. The full record, including that, is kept in the administration audit log.

When	What happened	Result	Details

Service overview

The whole deployment in one screen: accounts, conversations, queue, files and hardware.

Users

Every account on this deployment. Every change here is written to the audit log.

	User	Presence	Role	Status	Created	Last sign-in

Approvals

Accounts that passed the sign-up check and are waiting for a decision. Oldest first.

Approving lets the account use Q immediately — no new sign-in needed. Rejecting deletes it; the person can register again.

	User	Registered	Waiting

Conversations

Every topic on the service, with its owner.

	Title	Owner	Model	Updated

Uploaded files

What users sent to Q.

File contents are not shown and cannot be reached from here — only the operational record: who sent what, how big it was, and whether Q could read it.

	File	Type	Size	User	Extraction	Uploaded

Generated files

Documents, sheets and images Q produced for users.

	File	Format	Owner	Created

Queue

Requests currently generating or waiting for a slot.

Only a request that is still waiting can be cancelled. One that is already generating is holding a GPU slot and is left alone.

	Request	User	Status	Position	Created

System

CPU, memory, disk and GPU on the host running Q.

Model

Which model this deployment serves. Switching applies to every user.

Security & sign-up

Who may create an account, and how the three anti-spam defences are doing.

Sign-up policy

Changing this decides whether strangers can create accounts on this deployment, so it is recorded in the audit log.

Policy

Verification image (CAPTCHA)

Generated and checked by Q itself — no third party, no outbound request, no API key. Each challenge is single-use, short-lived and bound to the browser that asked for it.

Rate limits

Failed sign-in lockout

Branding

Everything this deployment is called and looks like — its marks, its name, its sign-in screen. One admin-controlled server setting per item, not a per-browser preference: normal users have no route that can change any of it, and every page reads all of it from the server on load.

Images

Uploads are checked before they are stored: a file that is not really an image, or is too small to draw the mark at, is refused with the reason. Anything larger than the display size is scaled down for you, keeping its proportions. Nothing is ever overwritten — every previous version stays available under History below.

Words

The strings the interface draws. These used to be written into the pages themselves, which is why a rebranded deployment still called itself Q on the sign-in screen.

Preview

The sign-in screen and the sidebar as a user will see them, drawn from what is saved right now.

History

Every branding change, newest first. Rolling back restores that version's images and text together — and is itself recorded, so it can be undone too.

Feedback

The thumbs up and down users left on Q's answers.

Rated	User	Rating	Model	Comment

Audit log

Every administrative action, append-only — including the ones that failed.

When	Admin	Action	Target	Resource	Result	Details

API keys

Programmatic credentials, one per integration. A key acts as its owner and can never reach this admin area. Every key issued, revoked or deleted is written to the audit log.

	Key	Owner	Can do	State	Created	Last used	Expires

Learning overview

What Q has been taught, where it came from, and what is still being read.

Verification

A verdict is evidence arithmetic, not an opinion: two independent sources have to agree before a claim is CONFIRMED, and INSUFFICIENT_EVIDENCE is an outcome, not a failure — it means Q looked and could not establish the claim, so nothing was promoted. How many people said something is never part of the sum.

Runners

The queue below is in memory. The durable record is the job table — a job interrupted by a restart is resumed from its row, not from this queue.

Latest jobs

Job	Source	Status	Progress	Started

Learning jobs

Every learning run, whether it was started from the chat or from here — these are the same records the chat's progress card reads.

	Job	Kind	Source	Command	Status	Stage	Progress	Counted	Started	Actions

Knowledge

What Q can retrieve, with where each piece came from.

ADMIN items are an administrator's own words. WEBSITE and FILE items are content Q read from a page or a document an administrator pointed it at — retrievable and attributed to that source, which is not the same as an administrator asserting it. Attributed is not verified. The Verified column says "checked" only for a row a verification CONFIRMED; everything else was read and stored, and a SUPERSEDED row is history that is never returned as the current answer.

	Subject	What Q learned	Source type	Source	Authority	Status	Verified	Version	Confidence	Learned	Details

User signals

Factual-sounding claims users made in chat, recorded as candidates.

How many people said it is not evidence. The Said by column is a count of informants and it carries weight zero in every verdict — a claim becomes promotable when independent external sources agree, never when users do.

	Claim	Class	Status	Used to answer?	Verdict	Promotion	Said by	Confidence	User	Recorded	Actions

Sources

Every origin Q has learned from, and when it last read each one.

A source is scored the first time it is used as evidence, not when it is first mentioned, so an empty trust score means never scored — which is not the same as untrusted. Trust is evidence weight and nothing else: it grants no permission, and content from a trusted source is still data, never instructions. Source trust opens the full breakdown.

	Source	Type	Identifier	Times learned	Last read	Trust	Why	Details

Resource review

Everything Q has learned from, and how much of it a person has actually checked.

Verification here is per claim. Opening a resource lists the statements Q took from it, each with its own provenance, audience, weight and verification state, and you verify the statements — not the source. Verifying marks a claim as checked by you: it does not change who may receive it and it does not change how much weight it carries. Those are separate operations on separate pages, each recorded on its own.

Resource	Type	Live claims	Reviewed by a person	Never checked	Authoritative	Global	Contested

	Claim	Audience	Weight	Checked	Freshness	Provenance	Decide

Automatic refresh

What Q re-reads on its own, when it last looked, and what it found.

A schedule decides when Q looks. It decides nothing about what is true: a change Q finds becomes a verification job, and knowledge changes only if that verification agrees. A cosmetic change — a timestamp, an advert, a reworded menu — creates no knowledge version at all. A source that stops responding is never treated as evidence that its knowledge was wrong; Q keeps the fact, marks the source unavailable, and eventually asks somewhere else. Only website and file sources can be refreshed: nothing here can reach a user's conversation.

Source	Policy	Status	Health	Last refresh	Next refresh	Last success	Last failure	Fails	Last change	Actions

Refresh history

Why Q changed — or did not change — its knowledge. Each row carries the hash before and after, how the change was classified and why, and the verification jobs it raised.

When	Source	Trigger	By	Outcome	Change	Why	Knowledge impact	Verification	Job

Knowledge health

How old each fact is, how healthy its source is, and when Q will look again.

Freshness is about age, not truth, and the threshold is different for different kinds of fact — a founding date and a team's current lineup do not go out of date at the same rate. A stale or expired fact is ranked lower and labelled with its age. It is never hidden, retired or deleted for being old.

Fact	Value	Kind	Freshness	Age	Verified	Last checked	Next check	Why	Details

Verification

Every verdict Q has reached, with the arithmetic that produced it.

Claim	About	Verdict	Why	Evidence	Sources	Weight for / against	Decided	Details

Evidence

What Q actually read, one row per source and location.

Every excerpt below was neutralised as text before it was stored, and none of it can change a verdict by what it says: stance is decided by comparing the claim's value with the value observed in the text, so a page containing "ignore previous instructions" is evidence about that page and nothing more. Trust at the time is a snapshot — it is what the source was worth when this was read, not its standing today.

Stance	Why	What the source said	Where	Observed value	Weight	Trust at the time	Read	Details

Conflicts

Questions Q holds more than one answer to.

Closing a conflict records which side won. It does not change what Q holds — setting a value is a separate, separately audited act under Knowledge versions, and keeping the two apart is what stops "resolve" from being an untraceable way to overwrite a fact. Every side stays on the record either way.

Question	Sides	Answers offered	Status	Outcome	Opened	Actions

Disagreements not yet recorded

Two sources giving different answers to the same question, with no conflict on file. Recording one does not retire either answer and does not decide anything — it puts the argument on the record so it can be resolved in the open.

Question	Answers held	Sources	Actions

Knowledge versions

What a fact used to say, kept exactly as it said it.

History is never deleted when a fact changes and it is never returned as the current answer. A superseded version is shown here labelled as history, and it cannot be made current by re-learning or re-verifying it — only a new version or an explicit override moves what Q holds. Re-reading an unchanged source raises the read count and creates no new version.

Version	Value	Status	Current?	Verified	Times re-read	Valid from	Valid until	Replaced by	Details

Source trust

What Q thinks of each source, and how it got there.

Trust is evidence weight, not permission, and not truth on its own: one highly trusted source does not confirm a claim, and no score anywhere grants a capability or lets a source's content act as an instruction. A score moves only by agreeing and disagreeing with evidence.

Source	Type	Trust	Made of	Why	Agreed	Disagreed	Grants	Details

Learning history

The audit trail of teaching — including the commands that were refused.

Read from the same append-only audit log as the Administration section, filtered to learning actions. There is no second history table.

When	Actor	Action	Resource	Result	Details

Capabilities

What the learning system does today and what it does not do yet.

Capability	Stage	Available	Attaches at

Deliberately not built

Not built	Why not

Overview

Profile

Security & devices

API keys

My topics

My files

Account activity

Service overview

Users

Approvals

Conversations

Uploaded files

Generated files

Queue

System

Model

Security & sign-up

Branding

Feedback

Audit log

API keys

Learning overview

Verification

Runners

Latest jobs

Learning jobs

Knowledge

User signals

Sources

Resource review

Automatic refresh

Refresh history

Knowledge health

Verification

Evidence

Conflicts

Disagreements not yet recorded

Knowledge versions

Source trust

Learning history

Capabilities

Deliberately not built

File types Q can learn from