Game Publisher Studios

New Run

Game Title

Stage

Steam ID

Co-signer

Unattended mode (--yes)

Context Documents

Lanes to run

Loading lanes…

No jobs yet.

Select a job or start a new run.

Completed Runs

Loading…

Recipes

Run a focused analysis recipe. Each recipe runs a curated subset of agents and produces a targeted output. Select a recipe, enter a game title, and submit.

Recipe Model profile

Game Title Stage

Context documents (PDFs, MDs, TXTs)

Job Progress

Log (0 lines)

Dispatch — Single Agent / Single Skill

Run any one agent or any one skill against any configured LLM profile, isolated in its own run directory. These hit POST /api/agents/{name}/run and POST /api/skills/{name}/run.

Run a single agent

Agent Model profile Slug Stage

Prompt

Context documents (PDFs, MDs, TXTs)

Run a single skill

Skill Model profile Slug Stage

Arguments (one per line)

Context documents (PDFs, MDs, TXTs)

Skills & Commands

Interactive (Claude Code CLI): Type slash commands in a claude session from the project root.
Headless (API): Full parallel runs via POST /api/parallel.

Primary Orchestrators

# Headless parallel run (recommended)
./jam.sh start --title "Game Title" --stage gold
./jam.sh start --title "Game Title" --stage gold --signoff "Name" --yes
./jam.sh start --title "Game Title" --stage gold --skip-chunks "E K"
./jam.sh retry banner-saga A # re-run one lane
./jam.sh resume banner-saga

# Interactive (in claude Code session)
/jam "Game Title"
/jam game-title gold # chunked, resume between sessions
/jam resume game-title
/slowjam "Game Title" # phase-by-phase with pauses

All Skills

Loading…

Configuration

API Server

Loading…

Data Connections (.connections.json)

Credentials come from environment variables — never stored in the file. Set missing vars in .env at the project root, then restart the API.

Loading…

Model Profiles (.jam.json → models)

Edit .jam.json → models.profiles to change or add profiles. Per-chunk overrides go in chunks.CHUNK_ID.model_profile.

Loading…

Multi-LLM Comparison (.claude/config/multijam.json)

Enable models by setting "enabled": true in multijam.json. Requires the relevant API key or a running local Ollama/LM Studio instance.

# Local models (free)
curl -fsSL https://ollama.ai/install.sh | sh && ollama serve
ollama pull gemma3:27b
/multijam "Game Title"