Set up local Ollama image generation for Marcus

sscoble commented

2026-04-29 03:18:03 +00:00

Owner

Goal: Enable Marcus to use local Ollama on MacBook Pro for image generation

Background:
Currently Marcus cannot generate images. Need to expose local Ollama instance (running on MacBook Pro) to the agent system for image generation capabilities.

Requirements:

Verify Ollama is installed and running on MacBook Pro
Confirm image generation models are available (e.g., llava, bakllava, or dedicated image models)
Determine network path for Marcus to reach the MacBook
Configure OpenClaw/agent system to use the local Ollama endpoint
Test image generation from Marcus

Notes:

MacBook Pro must be powered on and connected to network
Consider Tailscale or similar for secure remote access if needed
Image generation models may require significant RAM/GPU

Priority: Medium — would enable richer visual content for brand site and other projects

**Goal:** Enable Marcus to use local Ollama on MacBook Pro for image generation **Background:** Currently Marcus cannot generate images. Need to expose local Ollama instance (running on MacBook Pro) to the agent system for image generation capabilities. **Requirements:** - [ ] Verify Ollama is installed and running on MacBook Pro - [ ] Confirm image generation models are available (e.g., llava, bakllava, or dedicated image models) - [ ] Determine network path for Marcus to reach the MacBook - [ ] Configure OpenClaw/agent system to use the local Ollama endpoint - [ ] Test image generation from Marcus **Notes:** - MacBook Pro must be powered on and connected to network - Consider Tailscale or similar for secure remote access if needed - Image generation models may require significant RAM/GPU **Priority:** Medium — would enable richer visual content for brand site and other projects

sscoble added this to the 📋 MVP milestone 2026-04-29 03:18:03 +00:00

sscoble referenced this issue

2026-05-02 05:11:38 +00:00

Homepage hero has no background image or visual identity #50

agent-engineering commented

2026-05-05 22:06:56 +00:00

Collaborator

NOT STARTED - Local Ollama image generation not set up. Image generation for Marcus not present in codebase. This is outside brand-site scope and should be handled by marcus/core system.

sscoble commented

2026-05-25 02:04:07 +00:00

Author

Owner

This requires physical access to the MacBook Pro to verify Ollama is running, install image models, and configure network access for Marcus. @sscoble — when you have the MacBook, please:

Verify Ollama is running (ollama list)
Pull an image model (ollama pull llava or similar)
Determine the network path (Tailscale IP or local LAN address)
Configure the agent system endpoint
Test image generation from Marcus

Leaving this open for you to handle when the MacBook is available.

This requires physical access to the MacBook Pro to verify Ollama is running, install image models, and configure network access for Marcus. @sscoble — when you have the MacBook, please: 1. Verify Ollama is running (`ollama list`) 2. Pull an image model (`ollama pull llava` or similar) 3. Determine the network path (Tailscale IP or local LAN address) 4. Configure the agent system endpoint 5. Test image generation from Marcus Leaving this open for you to handle when the MacBook is available.

sscoble commented

2026-05-26 01:16:54 +00:00

Author

Owner

BLOCKED: MacBook Pro (scotts-macbook-pro-2) has been offline for 7 days per Tailscale. Cannot set up local Ollama without that machine being reachable. Scott needs to power on and connect the MacBook Pro to set up Ollama image generation.

sscoble commented

2026-06-02 04:46:09 +00:00

Author

Owner

🛠 MacBook Ollama runbook ready (2026-06-01)

CoS investigated. The blocker is the MacBook Pro itself — last seen on Tailscale 15 days ago. The setup work is straightforward but the hardware has to be online.

File: ~/vault/scott-inbox/ollama-macbook-runbook.md — 7-step runbook, 30-45 min cold / 5-10 min warm.

Quick summary:

curl -fsSL https://ollama.com/install.sh | sh
ollama pull llava bakllava llama3.2-vision (disk: ~30GB free)
Bind to Tailscale IP only: OLLAMA_HOST=$(tailscale ip -4):11434 ollama serve
Persist via ~/Library/LaunchAgents/com.ollama.ollama.plist (template in runbook)
Verify from n97-mini: curl http://<mac-ip>:11434/api/tags
Wire to ~/.hermes/config.yaml auxiliary.image_gen
Test from Marcus

Gotcha worth noting: LLaVA-style models do image description (vision), not image generation. For actual generation, the runbook includes the brew install stable-diffusion alternative. If Marcus needs to create images (not just analyze them), the SD path is what you want.

Real blocker: MacBook offline. Once it's back, this is a 30-45 min job.

Recommendation: Park this ticket as tech-dependency (MacBook availability), unblock when MacBook returns. Runbook file in scott-inbox/ is the work artifact.

— Chief of Staff

## 🛠 MacBook Ollama runbook ready (2026-06-01) CoS investigated. The blocker is the MacBook Pro itself — last seen on Tailscale 15 days ago. The setup work is straightforward but the hardware has to be online. **File:** `~/vault/scott-inbox/ollama-macbook-runbook.md` — 7-step runbook, 30-45 min cold / 5-10 min warm. **Quick summary:** 1. `curl -fsSL https://ollama.com/install.sh | sh` 2. `ollama pull llava bakllava llama3.2-vision` (disk: ~30GB free) 3. Bind to Tailscale IP only: `OLLAMA_HOST=$(tailscale ip -4):11434 ollama serve` 4. Persist via `~/Library/LaunchAgents/com.ollama.ollama.plist` (template in runbook) 5. Verify from n97-mini: `curl http://<mac-ip>:11434/api/tags` 6. Wire to `~/.hermes/config.yaml` auxiliary.image_gen 7. Test from Marcus **Gotcha worth noting:** LLaVA-style models do image **description** (vision), not image **generation**. For actual generation, the runbook includes the `brew install stable-diffusion` alternative. If Marcus needs to *create* images (not just analyze them), the SD path is what you want. **Real blocker:** MacBook offline. Once it's back, this is a 30-45 min job. **Recommendation:** Park this ticket as `tech-dependency` (MacBook availability), unblock when MacBook returns. Runbook file in `scott-inbox/` is the work artifact. — Chief of Staff

Rows
Columns

Set up local Ollama image generation for Marcus #26

🛠 MacBook Ollama runbook ready (2026-06-01)