Build it once. Deliver it to every client.
For agencies
Every client engagement starts by rebuilding the same plumbing: login, permissions, logging, storage. It eats your margin and no client pays to see it. Sinas gives each client an isolated instance with all of that already in place, so your team bills for the thinking instead of the auth layer. Sinas is an initiative of WeAreBrain, and this is how we use it ourselves.
Sound familiar?
If you deliver AI work for clients, you have probably said at least one of these. Here is what changes when you build on Sinas.
WITH SINAS
It ships with the platform, so your team starts on the part the client is actually paying for.
WITH SINAS
One isolated instance per client, each with its own keys, permissions and model provider.
WITH SINAS
Package what you build once and install it on the next client’s instance in minutes.
WITH SINAS
The instance and its configuration go with the client, or they self-host it. Open source, nothing held hostage.
Faster time to value for every client.
Every AI engagement quietly includes the same infrastructure work, and none of it is what the client pays to see. Sinas ships it with the platform, so your first client sees results sooner, and whatever you build can be installed for the next one in minutes.
From day one: what you no longer build
ACCESS AND TRUST
Authentication and API keys
OTP login, scoped keys, JWT refresh. Built in, on day one.
Permissions
Role-based access on every call, so each client team sees only their own data.
Execution logging
Every agent call, function run and query, tracked with timing and errors.
DATA AND MEMORY
Connected to client data
Connectors and reviewed queries into each client’s own databases and tools.
State and memory
Namespaced key-value storage, so conversations survive between sessions.
File handling
Upload, version, validate and search, without wiring up storage.
RUNNING THE WORK
Scheduling and webhooks
Cron and inbound HTTP, without another service to operate.
Running at scale
Queues, retries and large batches, so busy moments never drop work.
Safe code execution
Python functions run in a locked-down container, away from client systems.
On every next client: build it once, deploy it again
Package what you build
An agent, a function pack or a whole template becomes one configuration bundle.
Deploy it again
Upload it to the next client instance and adjust the parts that differ.
Keep it private
A package is a file you hold, not a listing you publish. Your library stays yours unless you choose to share it.
One instance per client.
A tier carries several instances, and each client gets their own. That maps cleanly onto how agencies work, and it makes the platform a small, predictable line on each engagement rather than a build cost.
HOW IT USUALLY RUNS
– One instance per client, so data, keys and permissions never mix
– Several applications on that one instance: support agent, internal tool, workflow
– Starter (€49/mo) includes 3 instances, Professional (€199/mo, coming soon) includes 8
– Bill it through as a pass-through cost, or fold it into the retainer
PARTNER TERMS
Running several clients?
If you are deploying instances for a portfolio of clients, talk to us before you buy them one at a time. We would rather set up terms that fit how you deliver: consolidated billing, volume arrangements, and a single point of contact for support across all of them.
Ready for your client’s security, legal and procurement.
The three reviews every engagement goes through, answered in one place. Send it to them, and anything it does not answer we will answer in writing.
Security
Licence
Handover
What a security reviewer looks for first, and what they will find.
Every call is checked
Role-based access on every call. An agent’s tools are re-checked against the caller’s permissions when they run, so it never does more than the person using it.
A full audit trail
Every request is logged with who made it, what it did and how long it took. Queued work keeps the identity of whoever started it.
Their own credentials
Connectors can use per-user OAuth, so each person signs in with their own account and gets their own encrypted token.
Their choice of model
Pick the provider per agent: OpenAI, Anthropic, Mistral, or a local model through Ollama.




