Every AI model.
One beautiful home.
Your hardware.
Promptly is a self-hosted AI workspace for you and your people — chat with any model, research with live web search, build shared Workspaces, learn with an AI tutor, talk out loud, and automate the boring parts. All of it runs on your own box.
promptly.yourdomain.comOne picker, every model you care about — cloud or fully local.
A full AI workspace, not just a chat window
Promptly ships the features you'd normally stitch together from five different subscriptions — integrated, consistent, and private.
One question. A whole research team.
Ask something broad and the model fans it out with the run_agents tool: up to four sub-agents research in parallel — each with its own web search, page reading and private working context — and hand back one merged brief with de-duplicated citations.
- Genuinely concurrent — a comparison that would take a dozen sequential searches lands in one tool call.
- Your chat stays lean — each agent reads pages into its own context and returns only a bounded digest, so raw page text never floods the conversation.
- Watch it work — every agent reports live progress in the chat's Tool Activity card, exactly like below.
Tool ActivityChats, notes, canvas, boards & sheets — one shared brain
Everything inside a Workspace is indexed into the same retrieval layer. Ask a chat about your notes, your board, or a cell in your sheet, and it just knows. Click through the sidebar — the board really drags, and any item opens to the side.
Workspaces · Product LaunchRich collaborative docs with live cursors (Y.js CRDT under the hood), tables, callouts and task lists. The AI can propose edits directly into a note.
- Draft announcement post
- Confirm pricing page copy
- Record 90-second demo video
- Prep launch-day support rota
| A | B | C | D | |
|---|---|---|---|---|
| 1 | Channel | Leads | Conv % | Revenue |
| 2 | Newsletter | 1,240 | 4.1% | $18,420 |
| 3 | Product Hunt | 3,867 | 2.3% | $31,050 |
| 4 | Referrals | 642 | 9.8% | $22,310 |
| 5 | Total | 5,749 | — | $71,780 |
Your AI, on a schedule
Build node-graph flows that run while you sleep: cron and webhook triggers, HTTP calls to any API with a built-in credentials vault, AI steps, branching logic — and outputs that write straight back into your Workspaces.
- Triggers your way — cron schedules, manual runs, or webhooks from CI, forms and monitors.
-
Credentials vault — API keys stored encrypted, referenced as
{{secret.NAME}}, never echoed back. - Durable runs — a dedicated worker queue keeps flows alive through redeploys, with full run history.
- Writes where you work — post results as board cards, notes, sheet rows or chat messages.
Flow · Morning competitor digest0 7 * * 1-5{{secret.RAPID_KEY}}changes > 0Pricing watch boardTalk to it. It talks back. Nothing leaves your network.
Dictate messages with self-hosted Whisper speech-to-text and have replies read aloud by Kokoro text-to-speech — both running as local containers beside the app. Full voice conversations without a single audio byte going to a third party.
- Dictation everywhere — tap the mic in any composer and speak your message.
- Read-aloud replies — natural local TTS on any assistant message, with a hands-free voice mode.
- Private by architecture — Whisper and Kokoro run on internal-only Docker networks; cloud STT is strictly opt-in.
Voice modeTap the orb to start the voice demo
A real drive, wired into every conversation
Folders, starring, trash with undo, full-text search — plus everything the AI generates lands here automatically, linked back to the chat that created it. Mention any file in a message with @.
- Paranoid on purpose — magic-byte validation, EXIF/GPS stripping, filename sanitisation and per-user quotas on every upload.
- PDF, DOCX, images, data files — attach anything; text-only models even get AI-written captions for images via the vision relay.
- Findable forever — Recent, Starred, Shared and Trash views plus search across your whole library.
Files · My DrivePlus everything else you’d expect
No add-ons, no per-seat upsells — the whole toolkit ships in the same one-command install.
Deep Research
An agentic pipeline that decomposes your question, runs parallel web searches, reads the actual pages, checks for gaps, and streams back a structured, cited report.
Code Interpreter
Models write and run real Python in a network-isolated sandbox. Attach a CSV, ask a question, get back charts and clean data — with the output files saved to your drive.
Generated artifacts
Ask for a PDF report, an image, or a data export and it lands in your Generated folder — linked back to the conversation that made it.
Memory that earns its keep
Promptly quietly captures durable facts across chats — pinned facts always apply, the rest are retrieved semantically. Capped, deduplicated, and fully yours to edit.
Web search, three ways
Self-hosted SearXNG out of the box — no API key needed. Prefer Brave or Tavily? Flip a switch. Per-conversation off / auto / on control.
Truly multi-user
Invite-only accounts for your family, team, or study group. Share chats read-only or invite collaborators right into the thread.
Context you can see
A live context-window meter on every chat, one-click compaction when it fills up, and truncation detection so a cut-off answer never slips past you.
Saved prompts
Your best prompts become slash commands. Type / in the composer and fire a template instantly.
Installable PWA
Add it to your phone's home screen for a real app feel — with push notifications when a long answer or scheduled task finishes while you're away.
Subchats
Chase a tangent in a floating throwaway side-chat without derailing the main thread. Get your answer, close it, carry on — nothing polluted.
Custom AI personas
Wrap any base model in a curated system prompt plus a knowledge library. "Support Bot on GPT-4o" shows up in everyone's picker, grounded in your docs.
Find anything, fast
Hit Ctrl+K and search every conversation you own or share — full-text with semantic re-ranking, snippets highlighted, one click to jump to the exact message.
Chat folders
Group your chats into collapsible folders — each with its own system prompt and default model that every chat inside quietly inherits.
Multi-user security and admin tools that take it seriously
Not a toy deployment. Promptly ships the auth, audit and observability you'd expect from a product you pay for — because your household or team deserves it.
Real MFA
Authenticator TOTP, email one-time codes, hashed backup codes and 30-day trusted devices. Admins can require MFA org-wide.
Invite-only accounts
No open signup. One-time invites, account lockout after repeated failures, and rate limiting on every sensitive endpoint.
Usage analytics
Tokens, messages and dollar cost by day, user and model — so you always know exactly what your API keys are spending.
Full audit trail
Every login, lockout, MFA attempt and rate-limit trip is recorded and filterable in the admin panel.
Live console
Structured JSON logs with request context, error grouping and stack traces — streamed right into the admin UI.
Model governance
Enable or disable individual models per provider, set defaults, and see OpenRouter privacy badges (training & retention policies) before you let a model near your data.
One command. Whole platform.
Clone the repo and run ./install.sh — it checks prerequisites, generates your secrets, and brings up the entire stack: app, Postgres with pgvector, Redis, local search, the code sandbox, voice services and the automation worker. Migrations run themselves. A first-run wizard in the browser creates your admin account.
-
Air-gap friendly — add
--with-ollamaand it all runs local: models, embeddings, the bundled SearXNG search and Whisper/Kokoro voice. Zero cloud, no API keys. -
Everything bind-mounted — your database, uploads and models live in one
./datafolder. Back it up, move it, own it. - Health-checked — a single endpoint pings Postgres, Redis and search so your uptime monitor always knows the truth.
./install.sh
./install.sh --with-ollama
./install.sh --no-search
0
models via OpenRouter alone
0
services in one compose file
0
command to deploy
0
bytes of telemetry sent anywhere
A calm, warm interface — day and night
No harsh blues, no clinical greys. Promptly's palette is warm paper and terracotta, tuned carefully for both themes. This whole page uses the app's exact design tokens — hit the toggle in the nav to feel it.
Own it, or let us run it
Self-host free forever on your own hardware — every feature, no seat limits. Or skip the ops entirely and let us run a private, managed instance for you.
Run the whole platform on your own box. Your data never leaves your network.
- Every feature — nothing gated
- Unlimited users on your hardware
- Bring your own model keys, or run fully local with Ollama
- One-command install & updates
- Community support
We host, update, back up and secure a private instance for you — the whole platform, zero ops.
- Everything in Self-hosted
- We handle hosting, updates & backups
- SSL, uptime monitoring & security patches
- Priority support
- Optional managed model billing — no keys to wrangle
Frequently asked
What does Promptly cost?
What do I need to run it?
./install.sh: it checks prerequisites, generates secrets, brings up the full stack, and a first-run wizard walks you through creating the admin account. A GPU is optional and only matters if you want fast local models.Can it run completely offline?
--with-ollama (or point it at an Ollama you already run) for chat models and embeddings; the SearXNG web search and Whisper + Kokoro voice containers are already local. In that setup, not a single request leaves your network.Which AI models can I use?
Is my data used to train anything?
How many people can use one install?
Your models. Your data. Your box.
Stop renting your AI workspace. Promptly gives you the whole platform — chat, research, Workspaces, tutoring, voice and automations — on hardware you control.