A roundtable, not a chatbot.
aiDex runs your prompt through up to five AI personas at once. They speak in turn, they reference each other, and they push back. The moderator decides who goes next based on the question: Strategist for direction, Tech Lead for feasibility, Skeptic for the gotchas.
Custom personas
Each agent is a model + a persona prompt that you write. Mix providers (Strategist on Claude, Skeptic on Gemini) and reuse the team across conversations.
Smart moderator
A cheap, fast model picks who answers next, detects when the team converges, and routes follow-up questions between agents instead of bouncing them back to you.
Inline @-mentions and replies
Agents @-mention each other to debate specific points. You can quote-reply to any message and the team picks up on that exact thread.
Consensus + disagreement cards
When the team aligns (or doesn't), the moderator surfaces a summary card with sentiment color, green for proceed, amber for caution, red for don't-do-it.
Suggested next questions
Stuck? The moderator proposes 2-4 single-click follow-up questions that move the conversation forward, pick one and reply with one tap.
Drop in a DOCX. Get a revised one back.
Upload a document and the team reads it, flags issues by section and type (factual / clarity / structural / style / risk), and consolidates duplicate findings into one accept-or-reject decision per real issue. When you're ready, the Editor + Reviewer pass generates a revised .docx that keeps the original's language and structure.
DOCX, Markdown, and plain text
Up to 25 MB per upload. We extract clean text for the team to read while preserving the original file for download.
Per-agent issue extraction
Each agent emits structured issues (section, type, suggested fix) that show up as accept/reject cards in the right-side panel.
Moderator-consolidated suggestions
When three agents flag the same paragraph, the moderator merges them into one recommendation so you only decide once.
Two-pass revision (Editor + Reviewer)
One agent applies your accepted edits to the document. A second verifies every accepted edit landed and flags any over-reach.
Document-language detection
We detect the document's language at upload time. The revision stays in that language even when your team is chatting in another.
Document library
Every upload and every revision lives in /documents, with revisions tree-grouped under their original. Download, browse, or delete.
You pay for orchestration, not model margin.
We charge for the orchestration layer, chat, teams, the moderator, documents, never for inference margin. On BYOK, the models run on your own provider key, so you see just two costs: your aiDex plan and your provider's bill, with nothing added on top. On Pro and Power, a monthly pool of AI Credits runs on our keys at the provider's wholesale rate, no markup, and once the pool runs out, anything beyond it is billed at provider cost plus a 15% service fee. No hidden third invoice either way. Prefer zero per-call cost? Point aiDex at an Ollama instance on your own machine, your hardware, nothing leaving your computer.
Five providers + local Ollama
OpenAI, Anthropic, Google, DeepSeek, and any Ollama-compatible endpoint, local on your laptop, on a homelab box, or behind a private tunnel. Mix cloud and local models inside the same team.
Envelope encryption at rest
Cloud keys are wrapped with AWS KMS before they hit the database. Plaintext lives only in server memory, briefly, during a call. Local-only teams skip the server-key path entirely.
Per-key spend tracking
Today's spend and last-used timestamp surface on every cloud key card. Local Ollama runs show up as $0.00, the way they should. Rotate a key and the next call uses the new one, no restart, no downtime.
Never wake up to a surprise bill.
Daily and monthly USD caps are hard stops. We email you when you cross 80% so you can adjust before a cap fires, and again at 100% when calls are blocked. Every call is tracked per provider, per feature, per conversation, drill in from any cell.
Adjustable spend caps
Default to $20/day and $200/month. Set them tighter or looser from Settings.
Top-spend dashboard
Usage page shows 30-day trend, spend by model, spend by feature (chat vs moderator vs documents), and the 10 most expensive conversations.
Cache savings line
Anthropic prompt caching kicks in automatically on long conversations. We surface what you've saved so you know it's working.
Native in English, Portuguese, and Spanish.
The UI, the moderator, the agents, the consensus cards, all native in all three languages. The moderator detects when an agent slips and translates back. Document revisions stay in the document's original language even when the team chat is in a different locale.
PT-BR and ES aren't 'translated EN'
Strings are first-class in every language. The moderator prompt has a language directive with locale-anchored exemplars so it doesn't default to English.
Language-aware revision
Chat in Portuguese, edit an English doc, the revision comes back in English. The team and the document don't have to share a language.
A full REST + SSE API.
Call any model (or your whole team) from scripts, CI pipelines, Slack bots, or other agents. OpenAI-compatible solo mode, plus native endpoints for Team, Compare, Judge, and Pipeline. Bearer-token auth, per-key rate limits, streaming with delta chunks.
OpenAI-compatible /v1/chat/completions
Drop-in replacement for the OpenAI endpoint. Existing client libraries work out of the box.
All 5 modes via API
Solo, Team, Compare, Judge, Pipeline, every mode available on the web, now callable programmatically.
What's next.
The things we're working on now that haven't shipped yet. Order isn't promised, we ship what's ready.
More providers
xAI Grok, Mistral, and Cohere are on the build list, same BYOK pattern, same per-call cost transparency. See the full slate (with "Coming soon" badges) in the aiDex index.
Voice mode
Talk to the team out loud. Each persona gets a distinct voice via real-time TTS, like a conference call where everyone in the room is an AI.
Presentation Studio
The team co-authors a .pptx the same way they co-edit a document today: outline pass, content pass, design pass, with accept/reject per slide.
Mobile apps
Native iOS and Android wrappers around the responsive web app, with push notifications for long-running document revisions and consensus events.
On the roadmap, tell us your use case and we'll prioritise it.