Create your account & sign in
Free to create, with a one-time grant of welcome credits so you can try real research right away.
Continue with Google
Ready immediately — no separate email verification step.
Email & password
“Create account,” 8+ char password — you're dropped straight into the agent.
Verify your email
Browse right away; verify before running metered work. Resend in-app.
Forgot password?
“Reset password” emails a one-hour link. (Google users sign in via Google.)
What you get on day one
A free plan plus a one-time grant of welcome credits. After signing in you land in the agent at /chat. Working with a team? See Organizations & workspaces.
Install Cortexa
Install the same Cortexa web application in its own window—sessions, workspaces, files, billing, and new features stay in sync automatically.
Chrome & Edge
Use the Install icon in the address bar or choose “Install Cortexa” from the browser menu.
Safari
On macOS choose File → Add to Dock. On iPhone or iPad use Share → Add to Home Screen.
Online research, privacy-safe offline behavior
The installed app launches directly into the workspace. Research sessions, organization data, APIs, files, and artifacts require a connection and are never stored in the offline cache. If the network drops, Cortexa shows a reconnect screen instead of stale private content.
Plans, payment & account lifecycle
Free, Plus, Pro, and Enterprise share the same research surface; paid plans add larger monthly credit pools and controlled overage.
Plus is the first paid step
Plus includes 750 monthly credits for individual researchers. Pro and Enterprise provide larger pools for heavier and team workloads. Manage plan changes, invoices, cancellations, and payment methods from Billing settings.
Usage is shown in credits
The workspace shows how many credits each completed turn consumed, while Billing shows the monthly credit balance and any paid overage. Raw provider run cost is an internal platform diagnostic, not a separate charge shown on subscription usage receipts.
What a credit is worth
Inside your monthly allowance a credit is metered at cost: one credit is one cent of model spend, with no markup added. A cheaper answer therefore consumes fewer credits. Overage past the allowance is the one place a margin applies.
Fast and Max
The control next to the send button chooses how much work a turn does. Max is the default: the strongest model with the full tool budget, for literature scans and cross-referenced synthesis. Fast runs a smaller model with fewer research steps, so it answers sooner and spends far fewer credits — good for a definition or a quick lookup, and worth re-running on Max before you rely on the answer. The choice applies to the next turn only, so you can switch at any point in a session.
Switching to Max mid-conversation
When you switch to Max after asking something on Fast, Cortexa does not take the earlier answers on trust. It re-establishes the load-bearing figures, citations and claims from the sources before building on them, and tells you plainly if something it finds contradicts what was said earlier. It opens with a short line so you know why the reply takes longer, then answers. This happens once after a switch rather than on every following turn. Your own messages, uploaded files and Library results are unaffected — those do not depend on which mode was answering.
Past-due and cancelled accounts
A past-due subscription remains visible in Billing with an update- payment action that opens the Stripe portal. Cancelling keeps the account and paid access through the current period, then moves the account to Free. Deleting an account is separate: deletion starts a recovery window and cancels active billing.
Recommended flow
The cleanest first run is narrow: ask one question, ground it with files or sources, then turn the useful result into an artifact.
1 · Start a session
Ask a narrow first question; attach any source files that shape it.
2 · Ask for evidence
Request mechanisms, evidence tables, trade-offs, open questions — with citations.
3 · Run experiments
Python cells for cleaning, statistics, charts, or model-ready transforms.
4 · Save what persists
Send important outputs to the Library to reuse them later.
5 · Package it
Once the session has enough context, generate an abstract — the written summary Cortexa produces from it — or a deck or dashboard.
A good first request
“Summarize the current evidence for X, cite primary sources, separate established findings from hypotheses, and list the next three analyses I should run.”
Set your research preferences
Tell Cortexa who you are once, and shape each answer as you go. The controls under the composer set output length and tone, with More controls for depth, format, citations and search breadth; they start from your saved defaults and anything you change applies to that conversation only, so a detailed write-up now doesn't bloat your next quick question. Settings → Research preferences holds who you are and the defaults every new session starts from.
Research identity
Your role, focus area, work setting, and region — the vocabulary, evidence standard, and guidelines it should assume.
Answer style
Length, technical depth, tone (including an academic register), citation density, and whether you want prose, bullets, or tables by default.
Per-conversation, not permanent
Changing a control in a session pins it to that session, marked with a dot. It stays available beside the composer once the conversation is running, applies from your next message, and comes back the same way if you reopen the thread. “Save as my default” is there when you did mean it permanently.
Your own notes
Up to 1,500 characters of free-form instruction: house style, preferred databases, how you want numbers reported.
Live preview
“What Cortexa will see” shows the exact block your settings compile to, before you save it.
Where it applies
Saved preferences are added to the system prompt on every agent run — in the workspace, in Slack, through the API, and over MCP. Abstract, presentation, and dashboard generation use their own templates and do not read them. Preferences are per person; there is no organization-wide version today.
Defaults stay out of the way
A new account has no personalization block at all — nothing is added to the prompt until you change a setting and save. “Reset to defaults” removes it again.
Safety rules come first
Preferences describe audience and formatting, so they cannot switch off citations or override the safety rules. Turning off soft disclaimers still leaves the research-not-medical-advice line on clinical answers.
Prompt patterns
Good prompts tell the agent what evidence matters, which files to inspect, and what shape the answer should take.
Ask for structure + citations
Name the sections you want and request primary sources.
Name the file to use
Point the agent at a specific dataset or uploaded file.
Set the depth
Pick Fast or Max next to the send button — Fast is a smaller model for quick lookups, Max is the full research run.
Refine with follow-ups
Critique, narrow, or convert the answer into an artifact.