Projects that keep context
Keep files, installed plugins and execution history together. Pick up the next task where the last one left off.
Local software. Lasting work.
Projects / Actions / RunsOrgSDK Platform
A local home
for your AI’s
actual work.
Your AI brings the reasoning.
OrgSDK gives its work somewhere to live.
Run code, keep project files and turn useful work into tested, reusable actions. Run it locally, connect your AI client, and stay in control.
Docker install · Your server · Your AI client
PROJECT
LocalFinance operations
ActionsFilesPluginsRuns
summarise-overduev1
// Action body · TypeScript
const { invoices, minimum } = defineParams(z.object({
invoices: z.array(z.object({
amount: z.number(), daysOverdue: z.number(),
})),
minimum: z.number(),
}));
const overdue = invoices.filter(
(invoice) => invoice.daysOverdue > 0
&& invoice.amount >= minimum
);
return {
count: overdue.length,
total: overdue.reduce(
(sum, invoice) => sum + invoice.amount, 0
),
};Test the edge cases.
Overdue invoices · Empty list · Minimum amount
Illustrative project and action body. Example data only.
From a one-off task to a repeatable operation.Built for work beyond the chat ↘
01 / Give the work a foundation
A persistent execution workspaceKeep the work.
Improve the process.
A good result shouldn’t disappear into a chat history. OrgSDK keeps the code, permissions and evidence together, so the next run starts from something you can review.
Actions worth running again
Turn useful code into saved actions with immutable versions and explicit input and output contracts.
Tests before trust
Exercise actions with isolated mocked tests. Check expected results and catch calls a task should never make.
Access with boundaries
Give each external client scoped MCP access to the projects and action versions it needs. Manage credentials and approvals in one place.
Your tools, your plugins
Install project-owned plugin releases to expose the capabilities your work needs. Keep execution separate from your AI client.
A history you can inspect
Review runs, their status and results in the web interface. Keep a durable record beyond a conversation.
02 / From intent to a tested action
External reasoning. Controlled execution.Your AI client.
A shared place to act.
Keep using your external AI client for conversation and reasoning. Connect it to OrgSDK through MCP, with explicit scopes for what it can do.
- 01
Connect a client
Create a project and grant an external client the operations and project access it needs.
- 02
Build and test
Execute TypeScript, inspect the result, then save useful code as an action with contracts and mocked tests.
- 03
Run a known version
Grant access to an immutable action version. Review approvals and inspect the resulting run history.
You choose the reasoning. OrgSDK manages the execution.
The local server owns authentication, permissions and persistence. Registered executors run code through scoped HTTP access, without database access.
03 / Make it local
Start on your own machineYour machine.
Your next operation.
Install Docker Engine 26+ with Compose and volume-subpath support. Clone the repository and start the server.
Open http://localhost:8080 to create your administrator and first project. Then start the managed executors to run code.
No model-provider account is required to set up the server. Connect your own external AI client when you’re ready.
Read the installation guide01 / Start the platform
git clone https://github.com/OrgSDK/OrgSDK.git
cd OrgSDK
./scripts/start.sh02 / Start the executors
docker compose --env-file .env.deployment \
--profile runner up -d --build --waitStartup builds the application, starts PostgreSQL and applies migrations. Keep the generated .env.deployment file private. The default server binds to all interfaces; see the guide for loopback-only access.
Source-available. Personal, non-business use and internal use by businesses with fewer than 15 staff are permitted without a license fee. Other use requires a commercial license. Read the license ↗