Platform
Agent200 is an SDK, a dashboard and server-side infrastructure that give your agent its external capabilities: connectors, managed OAuth, workflows, model-driven tool execution and unified billing. You keep the agent.
Your app
You own thisAgent200 SDK · one API key
Agent200
Provides and executesExternal services
Reached through connectors, with the end user's authorization
AI models
Selected by name, per request
Paid services
Consumed and billed through Agent200
Examples The services named in this diagram are examples. The full list of connectors, models and paid services is in the integrations catalog.
The problem
Reading a user's email, searching Slack and calling a search API each mean a provider integration, an authentication scheme, tool definitions and execution logic. That work repeats for every service, every agent and every team.
Every provider is its own project: its own credentials, API, authentication and bill.
Also yours to build and maintain
Your agent talks to Agent200. Agent200 talks to the providers on its behalf.
Agent200
agent200.run()
Handled by the platform
Your users still authorize access to their own connectors. Agent200 manages the OAuth flow without bypassing the provider’s authorization process.
The SDK
Add the SDK to the codebase you already have and authenticate with an Agent200 API key. That key covers the models, connectors and custom tools enabled for the project, with no separate provider keys to collect and rotate.
run()
submits a request: a model, instructions, an input and the tools the model is allowed to use. Agent200 executes it and returns the result to your code, where it becomes one step in whatever your agent does next.
Every user-specific operation is tied to an end-user identity, ensuring Agent200 uses only that user's authorized connectors. An API key alone never grants access to private accounts; users must authorize access themselves.
const result = await agent200
.user("user_123")
.run({
model: "openai/gpt-5.6-sol",
instructions: `
You are a productivity assistant.
Use the available tools when necessary.
`,
input: `
Retrieve the latest LinkedIn post from Dvir Ben-Aroya.
If today is Sunday, also prepare my brief.
`,
tools: ["brief", "linkedin"]
});| Parameter | In the example | Purpose |
|---|---|---|
| User identity | .user("user_123") |
Determines the end user whose authorized connections are used. |
| Model | model |
Selects the AI model that processes the request. |
| Instructions | instructions |
Defines the model's behavior for the request. |
| Input | input |
Supplies the actual task. |
| Tools | tools |
Defines which configured capabilities the model can use. |
Conceptual The public SDK interface is not final. Method signatures, supported languages, response types, streaming and error handling are still being decided, and user identity may be supplied differently, for example through the SDK constructor. Treat this example as an illustration of the model, not as reference documentation.
Model-driven execution
When you call run(), Agent200 invokes the model you selected, hands it the definitions of the tools you made available, executes whatever the model calls, and feeds the results back until the request is complete.
Your app
agent200.run()
model, instructions, input and tools: ["brief"]
Agent200 · model
AI model
Reads the task and the tool definitions, then chooses which tools to call, if any.
Tool definitions provided
brief · name, description, inputs
Repeats until complete
Each tool result returns to the model, which may reason further or call another tool.
Agent200 · tool execution
Calls: brief
Your app
Final response
Returned to your code, to use in the rest of your agent's flow.
Listing a tool in a request only makes it available. The model decides during execution whether the task calls for it: brief runs on Sunday, not because it appears in the array.
The model can search Slack, read Gmail and then answer, all within a single run(). Agent200 manages the exchange between the model and each tool.
One tool call can stand for a whole custom workflow. The model sees a capability and its inputs; the endpoints behind it stay an implementation detail.
Illustrative Every step in the diagram happens inside one run() call. Your code sees a request and a response; the tool sequence in between stays inside Agent200.
Models and paid services
Supported AI models and paid third-party services are reached through your Agent200 key. Agent200 holds the provider relationships, makes the underlying requests and accounts for the usage.
Same app · different requests
Set model on each run() call. Use one model for simple work and another for harder reasoning; the model you pick drives the tool-calling loop for that request. If your app already talks to an LLM provider, that can stay as it is. Agent200's execution applies when you call it.
You pick a supported model per request. The lineup grows as Agent200 adds providers to the catalog.
Web search through Tavily would normally mean another integration, another API key and another bill. Through Agent200 it is a capability your agent can call; you pay for it on your Agent200 account.
Tavily is one example. Each paid service Agent200 partners with is reached through the same key and billed the same way.
Agent200 does not currently support bringing your own provider keys. Supported models and services run on Agent200's provider relationships. OAuth authorization for your users' own accounts is separate, and still required.
Services
Connectors cover communication, email, files, CRM and developer tools. Search and AI models run on your Agent200 key. You enable what your agent needs, per project.
The dashboard
The dashboard is the control surface for your integration, not a replacement for your own app or agent interface. What you enable here is what your API key can reach.
Keep development, staging and production apart. Each environment can carry its own settings, integrations and credentials.
productivity-agent
An API key belongs to one environment. It reaches the connectors and workflows enabled there, and nothing from the environments beside it.
Create and manage the credentials your app authenticates with.
per project · per environment
Select external services and the endpoints your agent may use.
Slack · Gmail
Configure scopes and permitted access: search Gmail, but never send.
scopes · endpoints
Combine endpoints into reusable capabilities with a visual builder.
brief
Access supported models and paid services.
openai/gpt-5.6-sol · Tavily
View service consumption and charges in one place.
consumption · charges
Manage apps and their environments.
dev · staging · production
Illustrative Each row above is one area of the dashboard: configure it once and every run() request in that project and environment inherits the setting.
Clear boundaries
Infrastructure is easier to adopt when its edges are explicit. Want a brief every Sunday? Your scheduler calls run() every Sunday. Agent200 executes the request; it doesn't decide when it happens.
| Agent200 does | Agent200 does not currently |
|---|---|
| DoesProvides SDK access to supported services | Does not currentlyRequire you to replace your existing agent framework |
| DoesManages OAuth for supported connectors | Does not currentlyBypass provider authorization or permission requirements |
| DoesExecutes model-driven tool-calling loops | Does not currentlySchedule recurring agent runs: your app decides when to call run() |
| DoesSupports reusable custom workflows | Does not currentlyRequire you to implement every underlying API request |
| DoesProvides unified model and paid-service billing | Does not currentlySupport bring-your-own provider API keys |
| DoesManages user-specific external connections | Does not currentlyTreat every end user as sharing the same credentials |
| DoesRetrieves and processes external API data | Does not currentlyPersistently index retrieved connector data as a product feature |
These boundaries describe the current product direction. They are not permanent technical limits, and they are not commitments about future functionality.
Integrate the SDK into the app you already have, and let Agent200 provide and execute its external capabilities.