Hand off the work.
Keep the proof.
One workspace for every AI model, premium or free. Chat, build software, write real documents and run agent teams, connected to Google Workspace and any tool you use, on our cloud or on your own servers.
Base fee + credits at cost · Your files stay yours
End to end, in one place
Chat, code, documents, agents, routines and connectors live in one workspace. No copying context between five tools, no five bills.
Fewer tokens for the same work
The right model for each step, documents read once into text, cached context and compaction keep every run lean.*
Of your data at Madav, if you choose
Run the whole stack on your own servers with your own models. Only sign-in and billing talk to our cloud.
Say what you need.
Watch it work. Keep everything.
Describe the outcome
In plain words, with your files, folders or a Google Doc attached. Madav picks a model that fits the task and your budget.
Watch it work, stay in charge
A visible plan and live narration. Every edit, command or email waits for your yes, or runs on the autonomy level you set.
Keep the proof
What changed, what was checked, what it cost. Real files in your folder or Drive, and a record of every run.
Everything AI can do for you,
without the tab juggling.
Madav replaces the chat app, the coding agent, the document tool, the automation tool and the model router you would otherwise stitch together.
Chat and build on any model
Ask a question or hand over a whole folder or repository. Madav plans, edits, runs the checks and shows you the evidence before it says done.
Specialists you set up once
Describe an agent in your own words. Call it with @ in any session, or put several to work as a team that hands work down the line.
Work that runs on a clock
Daily summaries, weekly reports, watchdogs. Routines run in the cloud after you close the laptop, read-only unless you allow more.
Real Word, Excel, PowerPoint, PDF
Files you can open and send, read once into plain text before any model sees them, so long documents cost far less.
Recommend me, Model gym, Routing
Madav suggests a model for your work and budget, ranks models on your own runs, and never spends on a paid model without approval.
Skills, connectors, plugins, hooks
Teach Madav a task once as a skill. Plug in any connector. Everything it can do and reach sits in one page you control.
Your look, every screen
Light, dark and custom themes across web, desktop and phone, so Madav feels like your own tool.
Hire specialists in a sentence.
Put them to work as a team.
Describe an agent in your own words and Madav builds it: its job, its model, the tools it may touch. Call one with @ in any session, or line several up so work flows from one to the next.
- Describe, don't configure · "Read every new contract in Drive and list the renewal dates." Madav writes the agent, you review it.
- Any model per agent · A flagship for the researcher, a small local model for the contract reader. Each agent keeps its own.
- Only the tools it needs · Each agent gets its own connectors, read-only unless you allow changes.
- @ in any session · Type @code-reviewer and hand over the work without leaving the chat.
- Activities · Every run of every agent, with what it did, what it asked and what it cost.
Real files, not chat transcripts
Madav writes Word, Excel, PowerPoint and PDF files you can open, send and sign, and reads any document once into plain text so it costs a fraction to work with.
Teach it once. It remembers how.
Turn a way of working into a skill: your report format, your review checklist, your release steps. Add plugins and hooks to shape what every session does.
Premium or free.
Cloud or local.
One list.
The frontier models when the work is hard, open models when it isn't, and local models when nothing may leave the building. Switch per task, or let Madav choose.
- Premium · Claude on your own Anthropic key or cloud account; GPT, Gemini, DeepSeek and more on Madav credits or your key.
- Free and open · Qwen, Gemma, Llama, GPT-OSS, Mistral, DeepSeek-R1 and more, on your laptop or your own servers.
- At cost · A flat base fee, then credits at the provider's own price, cache reads included. Spend limits per person and per team.
The whole catalog
Every model Madav knows, with where it runs and what it is good at.
Connect it to everything.
Literally everything.
Madav speaks MCP, the open standard for AI tools. Add any of thousands of connectors, your company's own APIs, or a server you write in an afternoon. Each one is read-only until you allow changes.
Any MCP server
Google Workspace
Lives where your
team already works.
Connect Gmail, Drive, Docs, Sheets, Slides, Calendar and Meet once. Madav reads what it needs, writes real files back to Drive, and asks before anything leaves your outbox.
- Sheets → Docs → Gmail · "Every Monday, summarise last week's sales sheet into a Doc and draft the email to the team."
- Drive as memory · Point a session at a Drive folder; contracts, decks and notes are read once into text.
- Calendar and Meet · Prep notes before each meeting, action items after, filed where your team looks.
- Admin-set for the whole org · Connectors set once by an admin, tokens kept in Madav's vault, never inside a session.
Set it once.
Read it Monday.
Routines are agents on a schedule. They run in Madav's cloud or on your servers, on the model you pick, with exactly the connectors you allow.
Up to 10x fewer
tokens per task.*
Most AI bills are paid for waste: the biggest model on easy steps, whole PDFs pasted again and again, the same context re-sent to five tools. Madav is built to not pay for any of it.
Recommend me and Routing send easy steps to small, cheap models and keep the flagship for the hard part.
Word, PDF, slides and web pages are turned into clean text before any model sees them, not sent as pictures again and again.
Repeated context is served from the provider's cache at a fraction of the price, and every token is metered so you see it.
Long sessions are compacted, not truncated, so the context meter stays low without losing the thread.
Mechanical work runs as deterministic tools, so the model reasons and the tools do the typing.
Your organisation's knowledge is loaded once for every session and every agent, never pasted by hand.
* Illustrative. Savings depend on the task and the models used; "up to 10x" compares Madav's routing, caching and document handling with running the same work through separate tools on one flagship model.
Your servers. Your models.
Our control plane.
Install Madav's signed package on your own servers or private cloud. Your code, documents and prompts never leave your network; only sign-in, billing and updates talk to Madav, and only outbound.
Data stays home
Code, files and prompts live on your servers and your storage. Nothing is trained on your data.
SSO and roles
Sign in with your company's identity provider. Admins set who can use which models, connectors and spend.
Audit trail
Every session, approval and tool call recorded in a write-once archive your security team can read.
Air-gapped models
Run open models on your own GPUs through Ollama or vLLM. Add premium models on your own keys only if you want them.
One context base for the org
Your rules, glossaries, allowed models and connectors loaded into every session on every server, kept in one place.
Why teams choose
Madav over a stack of tools
Start on the web.
Finish on your phone.
Web
The full workspace in your browser. Cloud sessions keep running after you close the tab.
Desktop
macOS, Windows and Linux: local models, your folders, a real terminal.
Your servers
The same app on your own address, for everyone in your company.
Madav Remote
Approve, steer and follow every session from your phone.
Sessions
One base fee. Every model at cost.
A flat base fee per user for the workspace, then credits for model use at exactly what the provider charges. No markup on tokens.
For one person: the full Madav, on the web and on your desktop. Early-bird pricing.
- The full workspace: chat, documents, agents, routines
- The desktop app: file system, terminal, tray routines
- Local, fully private models on your own hardware
- Time Machine for your real project folders
- Phone control and cloud sessions
- Premium models on credits at cost, or your own keys
Everything in Personal for each user, with one shared bill and one shared context.
- Everything in Personal, for every user
- One bill, shared credits at cost
- Spend limits per person and per team
- Shared agents, skills, connectors and routines
- Roles and an admin console
Madav on your own servers, with your models and your identity provider.
- Private AI server, data never leaves
- SSO, roles and audit archive
- Org-wide context base and connectors
- Onboarding and support
A creator's blog, shared forward.
The honest path from a first repository to a product shaped by hundreds of experiments, and every AI idea behind it in plain English. 66 free lessons on AI and Git.
Questions, answered.
How does pricing work?
Each user pays a flat base fee, which covers the workspace, agents, connectors and the web, desktop and phone apps. Model use is paid with credits at exactly the provider's price, input, output and cache included, with no markup. Local models on your own hardware use no credits, and your own keys bill you directly. Self-hosted Enterprise is priced with you.
Do I need an API key to start?
No. Starter models work the moment you sign in. Add your own keys whenever you like; they are kept sealed and never shown again.
Which models can I use?
74 models from 17 makers: Claude, GPT, Gemini, DeepSeek and other premium models, plus open models such as Qwen, Gemma, Llama, GPT-OSS and Mistral that run on your laptop or your servers. Claude runs on your own Anthropic key or cloud account.
How does Madav use up to 10x fewer tokens?
It sends each step to a model that fits it, reads documents once into text, reuses cached context, compacts long sessions and lets deterministic tools do mechanical work. The saving depends on the task.
Does it work with Google Workspace?
Yes. Connect Gmail, Drive, Docs, Sheets, Slides, Calendar and Meet as connectors. Each is read-only until you allow changes, and anything that sends or shares waits for your yes.
Can we run it on our own servers?
Yes. The self-host package runs sessions, connectors, the gateway and the web app inside your network, with your models. Only sign-in, billing and management talk to Madav's cloud, outbound only.
Is it safe to let it edit files or send email?
Every change is shown before it happens and logged after. Risky actions ask first, you choose the autonomy level, and spending on paid models never happens without approval.
What runs under the hood?
Claude runs through Claude Code on your own key. Every other model runs through OpenCode on Madav credits or your key. Madav adds the workspace, the tools, the metering and the keys around them.
Your work,
proven.
Every model, every tool, one workspace. On our cloud or your own servers.
Base fee + credits at cost · Your files stay yours