Skip to content

Instantly share code, notes, and snippets.

@jmanhype
Created January 19, 2026 00:32
Show Gist options
  • Select an option

  • Save jmanhype/3c8f6da05bf8b079ea933a87cd3fb119 to your computer and use it in GitHub Desktop.

Select an option

Save jmanhype/3c8f6da05bf8b079ea933a87cd3fb119 to your computer and use it in GitHub Desktop.
AI Factory Infrastructure - Complete Context Drop: 3090 PC, ComfyUI, Letta MAS, Ollama - IPs, URLs, Agent IDs, SSH (2026-01-18)

AI Factory Infrastructure - Complete Context Drop (2026-01-18)

Single source of truth for all infrastructure, credentials, and connection details

Network Topology

┌─────────────────────────────────────────────────────────────────────┐
│                        HOME NETWORK (192.168.1.x)                   │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│  ┌──────────────┐     ┌──────────────────────────────────────────┐ │
│  │   MacBook    │────▶│           3090 PC (AI Factory)           │ │
│  │ 192.168.1.191│ SSH │              192.168.1.143               │ │
│  └──────────────┘     │                                          │ │
│         │             │  ┌────────────┐  ┌────────────────────┐  │ │
│         │             │  │  ComfyUI   │  │   Letta Server     │  │ │
│         │             │  │   :8188    │  │      :8283         │  │ │
│         │             │  └────────────┘  └────────────────────┘  │ │
│         │             │  ┌────────────┐  ┌────────────────────┐  │ │
│         │             │  │Frame Server│  │   Ollama (3090)    │  │ │
│         │             │  │   :8189    │  │      :11434        │  │ │
│         │             │  └────────────┘  └────────────────────┘  │ │
│         │             └──────────────────────────────────────────┘ │
│         │                                                          │
│         │             ┌──────────────────────────────────────────┐ │
│         └────────────▶│        ZimaBoard (Proxmox)               │ │
│                       │           192.168.1.xxx                  │ │
│                       └──────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────┘

3090 PC (Primary AI Factory)

Connection Details

Property Value
Hostname 3090 PC
IP Address 192.168.1.143
SSH User straughter
SSH Command ssh straughter@192.168.1.143
OS Linux (Ubuntu/Debian)
GPU NVIDIA RTX 3090 (24GB VRAM)

SSH Access

# Connect to 3090
ssh straughter@192.168.1.143

# SCP file transfer
scp file.txt straughter@192.168.1.143:/home/straughter/

# Rsync videos
rsync -avz straughter@192.168.1.143:/home/straughter/ComfyUI/output/video/ /Users/speed/proxmox/videos/

Services on 3090 PC

ComfyUI (Video Generation)

Property Value
URL http://192.168.1.143:8188
Web UI http://192.168.1.143:8188
API Endpoint http://192.168.1.143:8188/prompt
Queue Status http://192.168.1.143:8188/queue
History http://192.168.1.143:8188/history
Output Directory /home/straughter/ComfyUI/output/video/

API Examples:

# Check queue status
curl -s "http://192.168.1.143:8188/queue" | jq '{running: .queue_running | length, pending: .queue_pending | length}'

# Submit workflow
curl -X POST "http://192.168.1.143:8188/prompt" -H "Content-Type: application/json" -d @workflow.json

Frame Server (Video Listing)

Property Value
URL http://192.168.1.143:8189
List Videos http://192.168.1.143:8189/list_videos

API Examples:

# List recent videos
curl -s "http://192.168.1.143:8189/list_videos" | jq '.videos[:5]'

Letta Server (Multi-Agent System)

Property Value
URL http://192.168.1.143:8283
API Base http://192.168.1.143:8283/v1
Agents http://192.168.1.143:8283/v1/agents
Tools http://192.168.1.143:8283/v1/tools

API Examples:

# List all agents
curl -s "http://192.168.1.143:8283/v1/agents" | jq '.[].name'

# Send message to agent
curl -X POST "http://192.168.1.143:8283/v1/agents/{agent_id}/messages" \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "Your message"}]}'

# List agent tools
curl -s "http://192.168.1.143:8283/v1/agents/{agent_id}/tools" | jq '.[].name'

Ollama (Local LLM)

Property Value
URL http://192.168.1.143:11434
API http://192.168.1.143:11434/api
Models Qwen, Llama, etc.

API Examples:

# List models
curl -s "http://192.168.1.143:11434/api/tags" | jq '.models[].name'

# Generate completion
curl -X POST "http://192.168.1.143:11434/api/generate" \
  -d '{"model": "qwen2.5:32b", "prompt": "Hello"}'

Letta MAS Agent IDs

Video Production Crew

Agent ID Role
Director agent-22069f59-7a79-4890-bf4f-1f2a69696267 Creative direction, style injection
Writer agent-e565b3e8-4a59-440a-89ab-6c279d61cfb0 Prompt crafting
Cameraman agent-f939736a-46fc-4115-a584-0a8cf896212a Video generation via ComfyUI

Agent Tools

Cameraman Tools:

  • submit_video - Submit video generation job (frames param: 121 default, 210 max)
  • check_video_status - Check if video completed

Tool IDs:

  • submit_video: tool-4a5176d7-d18d-4ab6-b55c-ff8ccaa06e16

LTX-2 Video Generation Settings

Frame/Duration Limits (VRAM Tested)

Frames Duration Status
121 5.0 sec ✅ Default
180 7.5 sec ✅ Works
200 8.0 sec ✅ Works
210 8.7 sec MAX RELIABLE
215+ 9+ sec ❌ Fails (VRAM limit)

Video Output

Property Value
Resolution 1920x1088
FPS 24
Format MP4
Filename Pattern LTX-2_00XXX_.mp4
Model ltx-2-19b-distilled-fp8.safetensors
Upscaler ltx-2-spatial-upscaler-x2-1.0.safetensors

Local Paths (MacBook)

Purpose Path
Video Downloads /Users/speed/proxmox/videos/influencer_test/
Project Root /Users/speed/proxmox/videos/influencer_test/
Claude Plans /Users/speed/.claude/plans/

Quick Reference Commands

Server Health Check

# Full stack health check
ping -c 1 192.168.1.143 && \
curl -s "http://192.168.1.143:8188/queue" | jq '.' && \
curl -s "http://192.168.1.143:8283/v1/agents" | jq 'length' && \
curl -s "http://192.168.1.143:8189/list_videos" | jq '.videos | length'

Sync Latest Videos

rsync -avz straughter@192.168.1.143:/home/straughter/ComfyUI/output/video/ /Users/speed/proxmox/videos/influencer_test/

Check Video Duration

ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 /path/to/video.mp4

Send Message to Director Agent

curl -X POST "http://192.168.1.143:8283/v1/agents/agent-22069f59-7a79-4890-bf4f-1f2a69696267/messages" \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "Your instruction here"}]}'

Submit Video Generation

curl -X POST "http://192.168.1.143:8188/prompt" \
  -H "Content-Type: application/json" \
  -d @workflow.json

Current Visual Style Library Progress

Batch Style Videos Success
BATCH_001 Baseline 5
BATCH_002 Cinematic 5
BATCH_003 Noir 5
BATCH_004 Kubrick 5
BATCH_005 Spielberg 5
BATCH_006 Wes Anderson 5
BATCH_007 Tarantino 5
BATCH_008 UGC Cinematography 10 90%

Troubleshooting

Server Not Responding

# Check if reachable
ping 192.168.1.143

# SSH and check services
ssh straughter@192.168.1.143 "systemctl status comfyui letta"

VRAM Issues

  • Max reliable frame count: 210 frames
  • If video generation fails silently, reduce frames
  • Check nvidia-smi on server for VRAM usage

Letta Context Overflow

  • Use two-tool async pattern: submit_video() + check_video_status()
  • Avoid waiting for video in single tool call (180s timeout)

Last Updated: 2026-01-18 Source: Letta MAS Video Production Sessions

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment