Quickstart

This guide gets you from repository clone to a working local Workflow run as quickly as possible — fully offline, zero Docker, zero database setup.

---

Prerequisites

• Node.js 22 LTS (highly recommended; Node 20+ supported)
• pnpm (canonical package manager)
• (Optional but recommended) Ollama installed and running locally for zero-cost offline models.

---

Step 1: Install and Configure

Get the dependencies installed and set up the local file backend mode. Data will persist safely in browser IndexedDB (fully offline):

git clone https://github.com/conectlens/lenserfight.git
cd lenserfight
pnpm install
echo 'DATA_SOURCE=file' > .env.local

---

Step 2: Start the Web App

Spin up the local developer server. The composition compiles in seconds:

pnpm nx run web:serve

Open http://localhost:3000 in your browser. You are automatically signed in as Local Dev — no signup, credit card, or cloud accounts required.

---

Step 3: Create Your First Lens

Go to the Lenses library inside the dashboard and create a simple, versioned prompt template:

1. Click Create Lens.
2. Input a Title (e.g., Haiku Builder).
3. Write your versioned prompt:

   Write a 3-line haiku about [[topic]] in the style of a strict system architect.

4. Define topic as an input parameter in the parameters sidebar.
5. Click Publish Version.

---

Step 4: Create a Workflow DAG

Go to the Workflows library to orchestrate your first multi-agent Directed Acyclic Graph (DAG) pipeline:

1. Click Create Workflow.
2. Drag your newly published Haiku Builder Lens onto the canvas toolbar.
3. Link inputs together. For a first run, keep the pipeline small and linear.
4. Add a static execution value for topic (e.g., TypeScript).
5. Save your workflow.

---

Step 5: Execute and Observe

1. Click Run Workflow from the canvas control panel.
2. Select your provider (e.g., Ollama if running locally offline, or a BYOK key like OpenAI or Anthropic).
3. Click Execute Run and watch the live execution:
   • Node statuses will change in real time on the canvas.
   • Token generation counts and execution latency appear as they complete.
   • Read the final generated haiku output directly in the inspection sidebar!

---

Step 6: 🤝 Document and Share Your Workflow

LenserFight is built to support collaborative prompt design and experimentation. If you have successfully executed a workflow, you are welcome to share your setup:

• Take a Quick Capture: Capture a screenshot of your finished DAG canvas or record a brief GIF of the active execution nodes to document the layout.
• Document Your Local Model Setup: If you executed the workflow offline using Ollama or another local provider, note which model you ran (e.g., llama3.2) and observe its execution latency.
• Share with the Community: You can drop your run results, screenshots, or workflow DAG configurations in our GitHub Discussions to discuss optimization strategies with other developers. You can also propose adding your walkthrough or tutorial to our repository showcases by opening a Pull Request.

---

Optional: Try Direct Execution from the CLI

You can also run models and check environment health directly from the terminal:

# Verify environment requirements are green
pnpm setup:doctor

# Execute a quick offline prompt via local Ollama
node dist/apps/cli/main.js run exec --ollama --model llama3.2 --prompt "Explain workflow DAGs simply"

---

Known limitations in file-backend mode

The following features require a full Supabase instance or a specific feature flag and are not available in DATA_SOURCE=file mode:

Feature	Requirement
CRON scheduling	Supabase + Supabase pg_cron configured for workflow dispatch
Approval gates	Supabase (requires agents.* schema)
SSE run event replay	Supabase (requires lenses.workflow_run_events)
Marketplace (/marketplace)	Supabase (requires lenses.lenses with visibility)
BYOK cloud execution	Supabase + CHAINABIT_API_URL env var
Connector marketplace	Not yet implemented (preview)
Battles arena	Supabase + operator-approved cloud battles

See Known Preview Surfaces for the full list.

---

What to do next

• Run an Offline Battle with Ollama
• Configure Local Inference Engines (vLLM, llama.cpp)
• Create a Custom Lenser Agent
• How to Contribute a Translation