ewan's docs

bluesky-ollama is a Python bot that fetches recent posts from a source Bluesky account, feeds them to a locally-running Ollama model, and posts the AI-generated result to a destination account. It combines the Bluesky API integration from bluesky-markov with local LLM generation.

Features

Fetches and cleans posts from any source Bluesky account
Generates new posts via Ollama (default model: llama3.2)
Rate limiting that respects Bluesky's API quotas (5,000 points/hour, 35,000/day)
Content validation: rejects empty, placeholder, repetitive, or spam-like output
Dry-run mode for testing generation without publishing
Random posting intervals (30 minutes to 3 hours)

Setup

Requires Python 3 and Ollama running locally.

git clone git@github.com:ewanc26/bluesky-ollama
cd bluesky-ollama
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
ollama pull llama3.2

Create .env:

SOURCE_HANDLE=source.bsky.social
DESTINATION_HANDLE=destination.bsky.social
CHAR_LIMIT=280
SRC_APP_PASS=xxxx-xxxx-xxxx-xxxx
DST_APP_PASS=xxxx-xxxx-xxxx-xxxx
BSKY_HOST_URL=https://bsky.social
OLLAMA_MODEL=llama3.2

Usage

# Run normally
python src/main.py

# Dry run — generates but doesn't post
python src/main.py --dry-run

# Use a different model
python src/main.py -m mistral

# Combine
python src/main.py --dry-run -m llama3.3

Structure

src/main.py — Orchestration and scheduling
src/bsky_api.py — Bluesky API interactions
src/ollama_gen.py — Ollama post generation and content validation
src/clean.py — Text cleaning utilities
src/time_utils.py — Rate limiting and interval scheduling

Logs are written to log/general.log.

Licence

AGPLv3.