Palette
Mode
Or one accent
Surface
Saved in this browser. ⌘K has them too.
By what you are building
One-click SaaS
Open code, built and deployed in one click
Ask for what you need. It plans, you approve, it ships — into your own isolated cloud of servers, databases and storage. Then every system it runs can be operated by conversation.
An example of the Assistant at work. You ask “Add a Redis cache to @crm-staging”. It reads what is running, answers with a three-step plan, you approve it, and it reports: @crm-redis-stg is running · @crm-api-stg switched with no gap.
@vm-prod-1 to 8 vCPU on weekdays@crm-prod@crm-api back to yesterday’s image
Works with what
you already run
19
database engines
in 9 categories, with backups
42
ready-made listings
apps and databases, installed as a stack you own
15s
between setup checks
a new server is watched until it is ready
10min
billing reconciled
every server and bucket tagged to its owner
/how-it-works
The same four steps whether you are adding a cache, shipping a feature or importing a spreadsheet — and every step leaves something you can read: a plan, a pull request, a deploy, a tool call.
01
Say what you need in plain words. Point at things with @handles; run exact actions with /commands.
02
It reads the code and what is running, writes a PLAN.md and waits for you. Then a coding agent builds it in a sandbox, with tests and review.
03
Each project ships as a container on your servers — from a push to GitHub or one click. Health-checked before it takes over; the old one stays up if it fails.
04
The system publishes its MCP tools, kept as the context an agent reads. Every call is checked, recorded and — when it changes data — confirmed.
The whole platform
/platform
Each organization gets its own servers, networks, databases, storage and secrets — nothing shared with anyone else. Every piece below works today.
Public trafficInternal, by container nameEverything else on the network is denied
Stacks of containers on private networks, drawn the way they run. A deploy has to pass its health check before it takes over — if it doesn’t, the old one keeps serving.
14:00 · now running
8 vCPU · 16 GB
Real servers in our locations. Grow CPU and memory whenever you like — or give a server a different size at different times of day.
Nineteen engines, from Postgres to vectors. Passwords generated and kept as secrets; daily backups into your own bucket.
•••••••••••••• encrypted · rotated just now
Encrypted, versioned, given to a container only at deploy time. When a value moves on, you see who still runs the old one.
crm-exports.storage.talkingsuites.com
S3-compatible buckets with a web address of their own, files you can browse and share — and downloads that cost nothing.
Everything every container prints, searchable across the organization — with secrets masked on the way out.
Install the GitHub App on the repositories you choose. Every Dockerfile becomes a project; a push can deploy itself, pinned to the commit.
README read: POSTGRES_PASSWORD is required
Start from something that works: ready-made apps and databases, any public Docker image with its README read for you, and the MCP registry.
/systems
A system is a product’s own interface — usually an MCP server, in one of your containers or at an address. What it offers becomes the context an agent reads.
Acme CRM — contacts, companies, deals and files.
contacts.merge to confirmcontacts.mergeOne running thing, two handles.
Restarting a process and merging customer records are not the same risk — so they never sit behind the same permission.
restart · logs · shellbounces a processstacks.managecontacts.mergechanges customer data for goodsystems.use/tour
Every screen here is the product’s own design — the one the app is built from.
Ask about anything in your organization or tell it what to build. It keeps a checklist, runs coding tasks and sub-agents, and asks before anything irreversible.
Every stack is drawn as it runs: the public edge, the application, the services behind it, and the private network between them.
Ready-made systems, databases and services, each with a page that says exactly what it will put on your server before you install it.
Each system’s MCP tools with what they take and return, whether they read, write or destroy, and how much they are used.
A harness receives every alert and decides who handles it — each specialist with its own grants, a person when production data is at risk.
A run is a trace: who acted, which tool was called, what waited for a person and what it cost — down to the coding round that failed.
/agents
An agent is what it reads, what it may touch and what wakes it: AGENT.md, its skills and rules, a model, the tools and accounts it is granted, and its triggers. It versions like code, because it is files.
A triggered agent drafts and waits: anything outbound is a person’s decision. And it can call only what it was granted, tool by tool — here, two tools of @crm-agentsystemproductionAcme CRMThe CRM’s own interface — the MCP tools an agent can call. Its data, not its process.14 tools · 4 read, 9 write, 1 destructivepermission: systems.use.
/trust
Safety here is a set of mechanisms you can inspect, not a badge: what the platform will not do on its own, and what it writes down when it does.
Servers, networks, databases, buckets and secrets belong to one organization. Nothing is shared with anyone else.
A tool that changes data asks first; one that destroys it asks for its name typed back. Every call lands in a ledger.
Encrypted and versioned, handed to a container only at deploy time — over stdin, never in a command or a log line.
Every create shows the exact Docker commands it will run, with secrets by name, before anything is made.
The new container must pass its health check before the old one goes. If it fails, the old one is still serving.
Passkeys, two-factor, single sign-on with OIDC or SAML 2.0, verified email domains — and you can require them.
An audit log of every change, filterable and exportable as CSV — including what staff did on your behalf.
Every server, volume and bucket is tagged to the record that owns it and reconciled every ten minutes.
Coding starts from a plan you approved. An agent never interrupts you: it asks its lead, then the assistant, and only then you — one question, with a default.
/pricing
No platform fee and no price per seat. Servers, storage and sandboxes are billed by what they use, and every price is shown before you create anything.
No platform fee. Pay only for what you run.
$0a month — then only what runs
Request early accessFor companies with security, compliance and scale needs.
Customa contract, set up with you
Talk to us/faq
Something else? Write to [email protected] or use the contact page — a person answers.
One place to build, deploy and operate software. Your organization gets its own isolated cloud — servers, containers, databases, storage and secrets — and every system you run there publishes an interface that agents can use. The goal is that you can do all of it by talking to it.
Servers, deployments, databases, object storage, secrets, workspaces connected to GitHub, the marketplace, systems that speak MCP, the tool registry, logs and usage — with domains, connectors, sandboxes, the assistant and the agents you build yourself.
TalkingSuites is in early access. Request access, tell us what you would like to run first, and we write back from [email protected].
On servers created for your organization in our locations — you pick the region, the size and the operating system, and you can change the size later. Nothing you run shares a machine or a network with another organization.
No. You choose what to run and the platform writes the configuration. If you do know Docker, every screen shows the exact commands a deploy will run before it runs them, and the names on the server match the names on the page.
A system is a product’s own interface — usually an MCP server, in one of your containers or at an address. Publishing one asks it what it offers: its tools, what each takes and returns, whether it only reads or changes things. That becomes the context an agent reads, and every call to it is checked and recorded.
Yes. Install the GitHub App on the repositories you choose; every Dockerfile in them becomes a project you can deploy. Each workspace keeps its environments — variables and secrets for production, staging and development — and a push can deploy itself, pinned to exactly the commit that was pushed.
The Free plan costs nothing; you pay for what you run. Servers and volumes are charged by the hour they exist — never more than their monthly price — and object storage by what it holds; downloads are free. There is no per-seat price. The assistant and agents run on prepaid AI credits, so they are never billed after the fact. Enterprise is a contract.
Each organization is isolated. Secrets are encrypted, versioned and only handed to a container at deploy time. Anything that changes data asks first and anything destructive asks for its name typed back. Sign-in supports passkeys, two-factor and single sign-on, and every change is in the audit log.
/start
Request early access and tell us what you would run first — or talk to us about Enterprise.