Pi Fusion

A multi-model deliberation harness for the Pi Coding Agent. Pi Fusion takes any complex technical question and runs it through a structured deliberation pipeline -- parallel expert panels, comparative analysis, and grounded synthesis -- producing answers that are more thorough and balanced than any single model can achieve alone.

Inspired by the OpenRouter Fusion design pattern. Written in Node.js with native ES Modules, requires no build steps, and ships with presets for OpenCode Go and OpenCode Zen.

---

How It Works

5x Mode (Full Pipeline)

                  [User Query]
                       |
         +-------------+-------------+
         |             |             |
         v             v             v
    [Technical]    [Devil's]     [Systems]       Tier 1: Parallel Panel
     [Expert]     [Advocate]    [Thinker]        (3 models, run in parallel)
         |             |             |
         +-------------+-------------+
                       |
                       v
               [Deliberation]                    Tier 2: Judge
                  [Judge]                        (Structured JSON comparison)
                       |
                       v
                 [Synthesis]                     Tier 3: Synthesis
                   [Model]                       (Grounded final answer)
                       |
                       v
                [Final Answer]

Five LLM calls total. Each panel expert has a different system prompt persona. The Judge produces structured JSON (consensus, contradictions, partial_coverage, unique_insights, blind_spots). The Synthesis model resolves the contradictions and produces the final answer.

3x Mode (Lean Pipeline)

Three LLM calls instead of five. Two parallel experts (Technical + Devil's Advocate) followed by one synthesizer that absorbs the Judge and Systems Thinker roles. Roughly 40% cheaper than 5x mode while preserving the core deliberation benefit.

---

Presets

IDs below were checked on 2026-08-24 against OpenCode Go, GET https://opencode.ai/zen/go/v1/models, OpenCode Zen, and GET https://opencode.ai/zen/v1/models. Do not copy a Zen-only id into a Go preset.

Preset	Mode	Models	Catalog	Use case
GLM-5.3 Fusion	3x	All glm-5.3	OpenCode Go	Default. Latest GLM on Go.
Balanced	5x	kimi-k3, deepseek-v4-pro, glm-5.3	OpenCode Go	Three labs, still on the Go subscription.
High Quality	5x	kimi-k3, qwen3.8-max	OpenCode Go	Kimi + latest Qwen on Go (qwen3.8-max is not on Zen).
Quality / Frontier	5x	grok-4.6, gpt-5.6-luna, kimi-k3	OpenCode Zen	Newest Grok. grok-4.6 is not on Go.
Custom	5x	User-defined	Any	Full control over every model slot.

Role assignments:

• GLM Fusion. Every slot is glm-5.3 on Go. Zen still lists glm-5.2 as its newest GLM, so this preset stays on Go.
• Balanced. Technical Expert and Synthesis: kimi-k3. Devil's Advocate and Judge: deepseek-v4-pro. Systems Thinker: glm-5.3.
• High Quality. Technical Expert, Systems Thinker, and Synthesis: kimi-k3. Devil's Advocate and Judge: qwen3.8-max.
• Quality / Frontier. Technical Expert and Synthesis: grok-4.6. Devil's Advocate and Judge: gpt-5.6-luna. Systems Thinker: kimi-k3. Served from https://opencode.ai/zen/v1 with OPENCODE_API_KEY. If you must stay on Go, the equivalent Grok slot is grok-4.5, not 4.6.

Which catalog an id lives on

ID	OpenCode Go (/zen/go/v1)	OpenCode Zen (/zen/v1)	Notes
glm-5.3	yes	no	Newest GLM on Go. Zen's newest GLM is glm-5.2.
kimi-k3	yes	yes	Newest Kimi on both.
deepseek-v4-pro	yes	yes
deepseek-v4-flash	yes	yes	Default file-agent model.
qwen3.8-max	yes	no	High Quality stays on Go for this reason.
gpt-5.6-luna	yes	yes	Also on the public OpenAI API.
grok-4.5	yes	yes	Newest Grok on Go.
grok-4.6	no	yes	xAI id grok-4.6 (12 Aug 2026). Also https://api.x.ai/v1. Putting this on a Go 3x preset will 404.
gpt-5.6-sol	no	yes	Public OpenAI flagship. Used for the packaged openai provider, not for Go.

Go also lists grok-4.5, glm-5.2, kimi-k2.7-code, and others. Those still work; the presets pick the newest id that exists on that catalog.

---

Getting Started

Install as a Pi Extension

pi install git:github.com/QuarkOS/Pi-Fusion.git

Or from npm:

pi install npm:@quarkos/pi-fusion

The Pi extension uses @earendil-works/pi-ai, which Pi already provides. The standalone CLI does not need that package.

Set Your API Key

The default presets use OpenCode Go. Set the API key in your environment:

Windows (PowerShell):

$env:OC_GO_CC_API_KEY = "sk-opencode-..."

Linux / macOS:

export OC_GO_CC_API_KEY="sk-opencode-..."

The Quality / Frontier preset uses OpenCode Zen (OPENCODE_API_KEY) because grok-4.6 is not on the Go catalog. The same OpenCode account key often works for both endpoints.

If no OpenCode key is found, the client falls back to OPENAI_API_KEY and standard OpenAI endpoints. Pi Fusion also reads keys from Pi's own auth.json if you have connected a provider through Pi.

Choose a Preset

Inside Pi, run /fusion-config to select a preset interactively. From the CLI, run:

npx @quarkos/pi-fusion --setup

---

Usage

Inside Pi

Once installed, two features are available:

• /fusion <prompt> -- Runs the full deliberation pipeline and streams the synthesized answer into your Pi session.
• deliberate tool -- Available to the Pi agent itself. When solving complex tasks, the agent can call this tool to get a multi-model deliberation on a sub-problem.

As a Model Provider

Pi Fusion registers itself as a model provider called fusion. You can select it as your active model in Pi's model picker, and every message you send will go through the deliberation pipeline automatically. The synthesis streams token-by-token, and any generated code is routed through a file-agent that saves files to disk using Pi's write tool.

Command Line

Submit a one-off query:

npx @quarkos/pi-fusion "Explain the tradeoffs between microservices and monoliths"

Interactive REPL mode:

npx @quarkos/pi-fusion --interactive

Verbose mode (shows individual panel responses):

npx @quarkos/pi-fusion "Write a thread-safe singleton in Go" --verbose

Override specific model slots:

npx @quarkos/pi-fusion "Test query" --models "judge=glm-5.3,synthesis=deepseek-v4-pro"

---

Architecture

index.js              Pi extension entry point (provider, commands, tools)
bin/pi-harness.js     Standalone CLI with setup wizard
lib/api.js            OpenAI-compatible streaming API client with retry logic
lib/config.js         Packaged default config (shared by CLI and Pi entry)
lib/presets.js        Go / Zen / OpenAI model IDs and preset maps
lib/deliberation.js   Deliberation orchestrator (3x and 5x modes)
lib/ui.js             Terminal formatting for CLI output

The API client streams all responses to prevent gateway timeouts, accumulates token usage across calls, handles Kimi-specific temperature constraints (temperature: 1.0), and retries transient network errors with exponential backoff. The file-agent step after synthesis uses deepseek-v4-flash (cheap, near-unlimited on OpenCode Go) with tool-calling to decide what files to save.

---

Configuration

Pi Fusion looks for a pi-harness.config.json in your working directory. If none exists, it uses the packaged default: 3x GLM-5.3 Fusion on OpenCode Go (https://opencode.ai/zen/go/v1). That file also includes an OpenCode Zen provider (https://opencode.ai/zen/v1, grok-4.6) and a public OpenAI provider (https://api.openai.com/v1, gpt-5.6-sol). The config file is generated automatically when you run the setup wizard.

Key fields:

Field	Description
provider	Which provider to use (opencode-go, opencode-zen, openai)
mode	3x (lean) or 5x (full pipeline)
providers.<name>.defaultModels	Model IDs for each pipeline slot
panel.<role>.systemPrompt	Custom system prompts per expert
judge.systemPrompt	Judge comparison instructions
synthesis.systemPrompt	Final synthesis instructions
fileAgentModel	Model used for the file-saving step (default: deepseek-v4-flash)

---

Development

git clone https://github.com/QuarkOS/Pi-Fusion.git
cd Pi-Fusion
npm install

npm test                  # Offline unit tests (no API keys)
npm run test:stream       # Stream protocol tests (offline, no pi-ai)
npm run test:live         # Live API deliberation (needs OC_GO_CC_API_KEY)

---

License

MIT