Introducing Agent Hosting

Introducing Agent Hosting

2025-12-19 · TheCorporation Team

Most agent platforms have a pricing problem. They charge per execution, per seat, or per agent — on top of the underlying LLM costs. You pay for the platform and you pay for the intelligence. For corporate operations that run continuously — scanning deadlines, reconciling accounts, monitoring compliance — those costs compound into a second governance overhead, which is precisely the problem you were trying to eliminate.

Today we’re launching agent hosting that works differently.

The model

You bring your own API key — Anthropic, OpenAI, or OpenRouter. You pay the LLM provider directly for the tokens your agents consume. TheCorporation charges nothing on top. No platform fee. No per-agent fee. No per-execution fee. No markup on tokens.

A typical compliance scan — checking all pending deadlines, verifying good standing, flagging upcoming obligations — costs a few cents. A document generation run costs less. A full governance audit across formation, equity, compliance, and treasury costs roughly what a cup of coffee costs. These are LLM token costs, paid at the provider’s published rate.

The agent runtime itself — the sandboxing, the scheduling, the tool execution, the policy gate — is infrastructure we provide at no cost, the same way we provide the rest of TheCorporation at no cost.

The runtime

Each agent runs in a sandboxed container with its own budget, scope, and tool configuration. The sandbox is the governance boundary — an agent can only access what its scopes permit, can only spend what its budget allows, and can only call tools that have been explicitly configured.

Three types of tools are available:

Built-in tools. Code execution, file I/O, and the full suite of TheCorporation governance operations — entity formation, cap table management, compliance scanning, document generation, tax preparation, banking operations, contract management.

HTTP tools. Call any external API. Agents that need to interact with government filing systems, banking APIs, or third-party services configure HTTP tools with the appropriate endpoints and credentials.

MCP tools. Connect MCP-compatible servers for extended capabilities. If you’re already running MCP servers for other parts of your stack, your corporate agents can use them too.

Controls

Autonomy without constraints is just chaos. Every agent operates under three layers of control:

Budget limits. Per-agent and per-execution spending caps, denominated in tokens. An agent that hits its budget stops. It doesn’t quietly overspend — it stops and reports what it accomplished and what it couldn’t finish.

Scope boundaries. The same scope system that governs all of TheCorporation. An agent with compliance_write can file compliance documents. It cannot touch the cap table. Scopes are enforced at the infrastructure level, not by the agent’s own judgment.

Approval workflows. Actions that exceed the agent’s autonomous authority — equity issuance, large treasury movements, bylaw modifications — automatically route to the appropriate human approver. The agent doesn’t decide what requires approval. The policy gate does.

Triggers

Agents can be invoked three ways:

Messages. Send a natural-language instruction and the agent executes it. Good for one-off tasks: “Generate the Q1 board report” or “Check if our registered agent appointment needs renewal.”

Cron schedules. Set a recurring schedule and the agent runs automatically. Good for continuous operations: compliance scanning every morning, treasury reconciliation every week, deadline monitoring every day.

Domain events. Trigger agents in response to changes in entity state. Good for reactive workflows: when new equity is issued, automatically update the 409A valuation queue; when a compliance deadline is created, schedule the filing agent.

Get started

Agent hosting is live. Configure your API key, define your first agent, and deploy. The runtime handles scheduling, sandboxing, tool execution, and audit logging. You handle the decisions that require a human.

The cost of running your corporate agents should be the cost of the intelligence they consume — nothing more.