# Kalama agent API

Kalama is Jason Stillerman’s consulting company. This page is its living document. Its current copy is a placeholder.

## Read

GET https://kalama.dev/api/document returns the published full markdown, versionId, reviewedWeek, currentWeek, and needsWeeklyReview. GET /llms.txt gives AI readers a concise summary. Drafts and history are private.

## Propose a complete document

Jason mints a named token under Owner → Agents. Keep the token in a server-side environment variable, outside prompts, browser code, source files, and logs. Token identity controls authorship. Never claim to be Jason.

POST https://kalama.dev/api/agent/drafts
Authorization: Bearer <KALAMA_AGENT_TOKEN>
Content-Type: application/json

Body:

```json
{
  "baseId": "current versionId from GET /api/document",
  "requestId": "a unique stable identifier for this proposal",
  "summary": "What changed and why",
  "markdown": "The complete proposed document"
}
```

A successful response is HTTP 201 with {id,status:"pending",published:false}. An exact retry with the same requestId and content returns the existing proposal. Changed content requires a new requestId. A stale baseId returns 409: read the current page and reconcile your changes before retrying. At most 20 new proposals per token per hour. No agent endpoint can publish, reject, mint tokens, or access another agent’s drafts.

GET /api/agent/drafts?id=<id> with the same token reads that proposal’s status: pending, published, or rejected. Published means Jason approved it at least once; GET /api/document tells you which version is live now. Revoked tokens stop working immediately.

## Content contract

Agent proposals can contain up to 50,000 bytes of UTF-8 markdown. No specific headings, phrases, notes, dates, or punctuation are required by the API. Jason’s own edits have no application-level content restrictions or length limit. Jason controls the document’s wording. Work from the current published text and his instructions. Never invent clients, revenue, or availability; ask Jason when facts are uncertain.

## Eve integration

The transport is ordinary HTTPS and works from an Eve tool without a new runtime or a GitHub write token. A reference helper lives in this repository at integrations/eve/kalama.ts. Bind KALAMA_AGENT_TOKEN as a secret in the agent deployment, and use its server-side submit/read helpers from an authorized tool. Do not put the token into model-visible arguments. Jason chooses which agent receives a token; creating this site does not install a tool or grant access in agent-platform.