Docs

KarpathyTalk is a small public network for positive, constructive human builders.

The core idea is simple: there are users and there are posts. A post is a markdown document with some social metadata around it. Everything else is derived from that. A "reply" is just a post with a parent. A quote post is just a post that points at another post. Timelines and profile feeds are queries over posts.

KarpathyTalk is read-open and write-human. There is currently no public write API.

Read Modes

KarpathyTalk exposes two read modes:

• JSON for code
• Markdown for humans and LLM agents

Use JSON when you want structured fields, pagination, and relationships.

Use markdown when you want the actual authored document directly.

Endpoints

Get a user:

GET https://karpathytalk.com/api/users/karpathy

{
  "id": 1,
  "username": "karpathy",
  "display_name": "Andrej",
  "avatar_url": "https://avatars.githubusercontent.com/u/241138?v=4",
  "github_url": "https://github.com/karpathy",
  "profile_url": "https://karpathytalk.com/user/karpathy",
  "feed_url": "https://karpathytalk.com/user/karpathy/feed.xml",
  "created_at": "2026-04-05T20:39:46Z",
  "follower_count": 0,
  "following_count": 0,
  "post_count": 3
}

Get a post:

GET https://karpathytalk.com/api/posts/4

{
  "post": {
    "id": 4,
    "url": "https://karpathytalk.com/posts/4",
    "author": {
      "id": 1,
      "username": "karpathy",
      "display_name": "Andrej",
      "avatar_url": "https://avatars.githubusercontent.com/u/241138?v=4",
      "profile_url": "https://karpathytalk.com/user/karpathy",
      "github_url": "https://github.com/karpathy",
      "created_at": "2026-04-05T20:39:46Z"
    },
    "content_markdown": "## nanochat jan7 miniseries\r\n\r\nI need to test out image upload so here is the January 7 [nanochat](https://github.com/karpathy/nanochat) miniseries. ...",
    "created_at": "2026-04-06T05:28:30Z",
    "edited_at": "2026-04-06T05:33:51Z",
    "like_count": 0,
    "repost_count": 0,
    "reply_count": 0,
    "revision_id": 5,
    "revision_number": 2,
    "revision_count": 2,
    "revision_created_at": "2026-04-06T05:28:30Z"
  }
}

That is basically the model. The user is straightforward. The post contains the authored markdown in content_markdown, plus timestamps, revision metadata, and social counts.

Two important details:

• posts are revisioned, so the canonical post URL stays stable while edits create immutable revisions
• content_markdown is the source of truth, and local markdown asset URLs are expanded to absolute https://karpathytalk.com/... URLs in JSON responses so the content is portable outside the site

Markdown Surfaces

Atomic documents are also available as markdown:

• GET https://karpathytalk.com/docs.md
• GET https://karpathytalk.com/posts/{id}/md
• GET https://karpathytalk.com/posts/{id}/raw
• GET https://karpathytalk.com/user/{username}/md

/posts/{id}/md returns a markdown document with frontmatter and body.

/posts/{id}/raw returns only the raw post body.

/user/{username}/md returns a lightweight markdown representation of a user.

JSON API

All JSON endpoints are public, read-only, and GET only.

There are two real objects:

• GET /api/users/{username}
• GET /api/posts/{id}

There is one canonical collection/query surface:

• GET /api/posts

/api/posts returns posts in reverse chronological order and supports a small filter language:

• author={username}
• has_parent=true|false
• parent_post_id={id}
• before={id}
• limit={n} up to 100

Examples:

• latest root posts:
  • /api/posts?has_parent=false
• child posts by a user:
  • /api/posts?author=karpathy&has_parent=true
• direct children of a post:
  • /api/posts?parent_post_id=4

RSS

Per-user RSS is available at:

• GET https://karpathytalk.com/user/{username}/feed.xml