KarpathyTalk Community Skill

Skill by ara.so — Daily 2026 Skills collection.

KarpathyTalk is a Go-based developer social network (Twitter × GitHub Gists) where posts are plain markdown, the social layer supports likes/reposts/follows/replies, and all data is openly accessible via JSON and markdown APIs — designed for both humans and LLM agents.

What It Does

GitHub OAuth sign-in (no new credentials)
Posts are GFM markdown with syntax-highlighted code blocks and image uploads
Social features: likes, reposts, quote posts, replies, follows
REST API returns JSON (for agents/code) or markdown (for humans)
Single Go binary + SQLite +
```
uploads/
```
directory — trivial to self-host
Built with: Go, SQLite, htmx, goldmark

Installation & Local Setup

1. Create a GitHub OAuth App

Go to GitHub → Settings → Developer settings → OAuth Apps → New OAuth App:

Field	Value
Application name	KarpathyTalk
Homepage URL	`http://localhost:8080`
Authorization callback URL	`http://localhost:8080/auth/callback`

Save the Client ID and Client Secret.

2. Clone & Build

bash

git clone https://github.com/karpathy/KarpathyTalk.git
cd KarpathyTalk
go build -o karpathytalk ./cmd/karpathytalk

3. Configure Environment

bash

export GITHUB_CLIENT_ID=$GITHUB_CLIENT_ID
export GITHUB_CLIENT_SECRET=$GITHUB_CLIENT_SECRET
export BASE_URL=http://localhost:8080   # optional, defaults to this

4. Run

bash

./karpathytalk
# or with options:
./karpathytalk -addr :9090 -db ./data/karpathytalk.db

Visit

http://localhost:8080

CLI Flags

-addr string    HTTP listen address (default ":8080")
-db string      SQLite database path (default "karpathytalk.db")

Environment Variables

Variable	Required	Default	Description
`GITHUB_CLIENT_ID`	✅	—	GitHub OAuth client ID
`GITHUB_CLIENT_SECRET`	✅	—	GitHub OAuth client secret
`BASE_URL`	❌	`http://localhost:8080`	Public URL of the deployed app

Deployment (Production)

Build & Copy

bash

# Build binary
go build -o karpathytalk ./cmd/karpathytalk

# Copy to server (adjust user/host)
scp karpathytalk schema.sql user@yourserver:~/karpathytalk/
scp -r templates static user@yourserver:~/karpathytalk/

Run on Server

bash

ssh user@yourserver
cd ~/karpathytalk
export GITHUB_CLIENT_ID=$GITHUB_CLIENT_ID
export GITHUB_CLIENT_SECRET=$GITHUB_CLIENT_SECRET
export BASE_URL=https://yourdomain.com
./karpathytalk -addr :8080

Caddy TLS (recommended)

caddyfile

yourdomain.com {
    reverse_proxy localhost:8080
}

nginx TLS

nginx

server {
    listen 443 ssl;
    server_name yourdomain.com;

    ssl_certificate     /etc/letsencrypt/live/yourdomain.com/fullchain.pem;
    ssl_certificate_key /etc/letsencrypt/live/yourdomain.com/privkey.pem;

    location / {
        proxy_pass http://localhost:8080;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }
}

systemd Service

ini

[Unit]
Description=KarpathyTalk
After=network.target

[Service]
WorkingDirectory=/home/deploy/karpathytalk
ExecStart=/home/deploy/karpathytalk/karpathytalk -addr :8080
Restart=always
Environment=GITHUB_CLIENT_ID=$GITHUB_CLIENT_ID
Environment=GITHUB_CLIENT_SECRET=$GITHUB_CLIENT_SECRET
Environment=BASE_URL=https://yourdomain.com

[Install]
WantedBy=multi-user.target

API Usage

All data is open — no auth required for reads. The API returns JSON for programmatic access and markdown for human/agent reading.

Fetch Posts as JSON

bash

# Timeline / recent posts
curl https://karpathytalk.com/api/posts

# Single post
curl https://karpathytalk.com/api/posts/{postID}

# User's posts
curl https://karpathytalk.com/api/users/{username}/posts

Fetch Posts as Markdown

bash

# Human/agent-readable markdown
curl https://karpathytalk.com/api/posts/{postID}.md

Go Agent Example — Read Timeline

package main

import (
    "encoding/json"
    "fmt"
    "io"
    "net/http"
)

type Post struct {
    ID        int64  `json:"id"`
    Username  string `json:"username"`
    Content   string `json:"content"`
    Likes     int    `json:"likes"`
    Reposts   int    `json:"reposts"`
    CreatedAt string `json:"created_at"`
}

func fetchTimeline(baseURL string) ([]Post, error) {
    resp, err := http.Get(baseURL + "/api/posts")
    if err != nil {
        return nil, err
    }
    defer resp.Body.Close()

    body, err := io.ReadAll(resp.Body)
    if err != nil {
        return nil, err
    }

    var posts []Post
    if err := json.Unmarshal(body, &posts); err != nil {
        return nil, err
    }
    return posts, nil
}

func main() {
    posts, err := fetchTimeline("https://karpathytalk.com")
    if err != nil {
        panic(err)
    }
    for _, p := range posts {
        fmt.Printf("[%s] %s (👍 %d)\n", p.Username, p.Content[:min(80, len(p.Content))], p.Likes)
    }
}

func min(a, b int) int {
    if a < b {
        return a
    }
    return b
}

Go Agent Example — Fetch User Posts as Markdown

package main

import (
    "fmt"
    "io"
    "net/http"
)

func fetchUserPostsMarkdown(baseURL, username string) (string, error) {
    url := fmt.Sprintf("%s/api/users/%s/posts.md", baseURL, username)
    resp, err := http.Get(url)
    if err != nil {
        return "", err
    }
    defer resp.Body.Close()
    body, err := io.ReadAll(resp.Body)
    return string(body), err
}

func main() {
    md, err := fetchUserPostsMarkdown("https://karpathytalk.com", "karpathy")
    if err != nil {
        panic(err)
    }
    fmt.Println(md)
}

Content Limits

Content Type	Max Length	Rate Limit
Posts	10,000 characters	30 per hour
Replies	5,000 characters	60 per hour
Images	5 MB	PNG/JPEG/GIF/WebP only

Database — Direct SQLite Access

The SQLite database is a single file. You can query it directly for analytics, backups, or migrations:

bash

# Open database
sqlite3 karpathytalk.db

# List tables
.tables

# Recent posts
SELECT username, substr(content, 1, 80), created_at
FROM posts
ORDER BY created_at DESC
LIMIT 20;

# Most liked posts
SELECT username, likes, substr(content, 1, 60)
FROM posts
ORDER BY likes DESC
LIMIT 10;

# User follower counts
SELECT username, COUNT(*) as followers
FROM follows
GROUP BY username
ORDER BY followers DESC;

Backup

bash

# Simple file copy (safe while running with WAL mode)
cp karpathytalk.db karpathytalk.db.backup

# Or use sqlite3 online backup
sqlite3 karpathytalk.db ".backup karpathytalk_backup.db"

Project Structure

KarpathyTalk/
├── cmd/
│   └── karpathytalk/      # main entrypoint
├── templates/             # HTML templates (htmx-powered)
├── static/                # CSS, JS assets
├── uploads/               # User image uploads
├── schema.sql             # SQLite schema
└── karpathytalk.db        # Database (created at runtime)

Common Patterns

Pattern: Agent That Monitors New Posts

package main

import (
    "encoding/json"
    "fmt"
    "net/http"
    "time"
)

type Post struct {
    ID        int64  `json:"id"`
    Username  string `json:"username"`
    Content   string `json:"content"`
    CreatedAt string `json:"created_at"`
}

func pollNewPosts(baseURL string, sinceID int64) ([]Post, error) {
    url := fmt.Sprintf("%s/api/posts?since_id=%d", baseURL, sinceID)
    resp, err := http.Get(url)
    if err != nil {
        return nil, err
    }
    defer resp.Body.Close()
    var posts []Post
    json.NewDecoder(resp.Body).Decode(&posts)
    return posts, nil
}

func main() {
    var lastSeenID int64 = 0
    for {
        posts, err := pollNewPosts("https://karpathytalk.com", lastSeenID)
        if err != nil {
            fmt.Println("Error:", err)
        } else {
            for _, p := range posts {
                fmt.Printf("New post by @%s: %s\n", p.Username, p.Content[:min(60, len(p.Content))])
                if p.ID > lastSeenID {
                    lastSeenID = p.ID
                }
            }
        }
        time.Sleep(30 * time.Second)
    }
}

func min(a, b int) int {
    if a < b { return a }
    return b
}

Pattern: Embedding KarpathyTalk Content in an LLM Prompt

package main

import (
    "fmt"
    "io"
    "net/http"
)

// Fetch markdown content to inject into an LLM system prompt or context window
func getContextForLLM(baseURL, username string) (string, error) {
    url := fmt.Sprintf("%s/api/users/%s/posts.md", baseURL, username)
    resp, err := http.Get(url)
    if err != nil {
        return "", err
    }
    defer resp.Body.Close()
    body, err := io.ReadAll(resp.Body)
    return string(body), err
}

func main() {
    context, _ := getContextForLLM("https://karpathytalk.com", "karpathy")
    systemPrompt := fmt.Sprintf(`You are a helpful assistant. Here are recent posts from the community:

%s

Answer questions about these posts.`, context)
    fmt.Println(systemPrompt)
}

Troubleshooting

Problem	Cause	Fix
`GITHUB_CLIENT_ID not set`	Missing env var	Export `GITHUB_CLIENT_ID` before running
OAuth callback fails	Callback URL mismatch	Ensure GitHub OAuth App callback URL exactly matches `BASE_URL/auth/callback`
Binary not found after build	Wrong output path	Run `go build -o karpathytalk ./cmd/karpathytalk` from repo root
`templates: no such file`	Missing templates dir	Run the binary from the repo root, or copy `templates/` and `static/` next to the binary
Database locked	Concurrent writers	SQLite WAL mode is set by default; avoid multiple binary instances against same `.db`
Images not serving	Missing `uploads/` dir	The directory is created automatically on first upload; check write permissions
Port already in use	Another process on 8080	Use `-addr :9090` to change the port

Check the App is Running

bash

curl -I http://localhost:8080/
# Expect: HTTP/1.1 200 OK

Rebuild After Code Changes

bash

go build -o karpathytalk ./cmd/karpathytalk && ./karpathytalk

Run Tests

bash

go test ./...

karpathytalk-community

NPX Install

Tags

SKILL.md Content

KarpathyTalk Community Skill

What It Does

Installation & Local Setup

1. Create a GitHub OAuth App

2. Clone & Build

3. Configure Environment

4. Run

CLI Flags

Environment Variables

Deployment (Production)

Build & Copy

Run on Server

Caddy TLS (recommended)

nginx TLS

systemd Service

API Usage

Fetch Posts as JSON

Fetch Posts as Markdown

Go Agent Example — Read Timeline

Go Agent Example — Fetch User Posts as Markdown

Content Limits

Database — Direct SQLite Access

Backup

Project Structure

Common Patterns

Pattern: Agent That Monitors New Posts

Pattern: Embedding KarpathyTalk Content in an LLM Prompt

Troubleshooting

Check the App is Running

Rebuild After Code Changes

Run Tests