64 healthy
For AI agencies
Build AI agents for every client.
Stackbone is the agent platform for AI agencies. One command scaffolds a production agent. Each client runs it isolated in their own cloud, and your team operates all of them from one place.
support-bot
bystackbone-demorunningv2.14.3created2026-08-19 10:04 active
datapath reachable p95 142 ms · checked 2 min ago
region eu-west-1 aws · byoc
connectors 5 linked Gmail expiring · Notion broken
secrets 6 stored oldest rotated 2026-08-09
75 with schema
540 documents · 2 collections · 22,530 chunks last ingest 2026-09-07 14:32
4waiting for a decision
13 done · 3 failed · 2 in flight · 2 interrupted
7b13d630…c4e failed
- Agent / Workflow
- invoice-parser
- Trigger
- job
- Started
- 14:18:40
- Duration
- 612 ms
- Tokens
- —
note input rejected: invoice_id missing from the job payload
Built for
How it works
A production agent in one command.
The stack comes provisioned, your coding agent writes the workflows on top of it, and the result runs inside the client’s own cloud.
npx stackbone init client-invoice-agentdatabase
RAG
model provider
integrations
observability
guardrails
01. One command, the whole stack.
The CLI scaffolds a workspace with the emulator, the database, RAG, the integrations, observability, guardrails and the model provider already mounted. You add the keys in Studio and the agent runs. Swap the database or the RAG in code when a client needs a different one, and Studio still drives it.
See local development02. Built for your coding agent.
The workspace ships the skills and the MCP server for Claude Code, Codex and Cursor. Your agent writes workflows against the SDK from the first prompt, inside the platform’s conventions.
See the coding-agent setupskillsthe house conventions
mcp serverthe platform, as tools
sdktyped workflows
read from the repo by
03. Into their cloud. Nothing leaves.
Each deployment is self-contained in the client’s own AWS, Azure or GCP account, their datacenter, or air-gapped. Provisioning ships with the platform, so infrastructure stays out of your proposal. Your team supervises from the control plane without breaking compliance, and every action lands in the audit log.
See how the perimeter holdsAWSalpha
Azurebeta
GCPgamma
Connectors
Their stack, plugged into the workflow.
Stackbone ships Gmail, Outlook, GitHub and the rest of the client’s stack with the authentication already declared. The broker handles OAuth, API keys and app-only credentials. Anything else runs as a custom connector, registered from an OpenAPI spec or an MCP server. You pick the ones the workflow needs and the coding agent writes the rest.
Any HTTP servicePoint a connector at a spec and Stackbone derives its operations without an adapter.
Any MCP serverRegister the server and its tools become steps the workflow can call.
And the rest of the catalog The list keeps growing. You register whatever else the client runs the same way: declare its authentication once, and its operations stay typed from there on. Explore every connector Features
The part the client lives with.
Shipping the workflow is the first week. Your team lives with these four for the years after.
Every step on the record.
A run keeps its steps: the model, the tokens, the reasoning, the prompt and its version. When a client asks why a workflow failed three months after go-live, your team opens the run instead of digging through logs. The client opens the same run.
See observability
Change a prompt, prove it first.
Your developer asks the coding agent. The client’s ops lead edits the prompt, or the mailbox it answers from, in Studio without touching code. Studio keeps every version, and the eval suite says whether the change still meets accuracy and cost.
See evals
A run outlives the request.
A workflow parks on an approval and waits for minutes or days, then resumes where it stopped. No connection stays open in the meantime, and the person who owns the decision gets the request.
See human-in-the-loop
The same box on your laptop.
stackbone dev boots your agents and durable workflows against a local Postgres, Redis and object store, and tunnels to Studio. The loop your team develops in is the one that ships to the client.
See local development Multi-tenant
Now do it for every client.
Each client gets their own tenant, deployed in their own cloud. Your team supervises all of them from one control plane with a single login. That is how the agency goes from its second client to its tenth.
client-alpha AWS eu-west-1 Isolated by construction
Each tenant runs in the client’s own cloud account. Their data stays inside their perimeter and never touches another client’s.
invoice-processing workflow running
contract-review workflow running
One control plane, every client
Every client’s runs, logs and approvals live in the same control plane, so your team drops the dashboard, the VPN and the credentials it kept per client.
weekly-report workflow running
Scale without new infra
Adding a client is adding a tenant. Your client list grows without your ops growing with it.
Pricing
One price per workspace. A margin you can quote.
Each client workspace has one fixed monthly price. You put a fixed fee in the proposal, and you know your margin before the client signs.
No usage surprises.
A workflow that runs ten thousand times costs the same as one that runs ten. Stackbone does not meter executions or count nodes and tasks.
Fixed fee in, fixed cost out.
Quote the client a monthly fee and know the margin before they sign. The invoice never depends on how busy their month was.
Model costs on their keys.
LLM usage runs on the API keys the client adds in Studio, inside their perimeter. Their spend stays theirs, and never lands on your invoice.
Bring your next client onto Stackbone.
Tell us what you’re building for them. On a call we take you from the first local run to the deployment in their cloud.