LOCAL-FIRST • FAST + DEEP • MCP READY

Research Workspace That Feels Premium, Runs Free, and Stays Grounded.

Argentic Search Lab is a local-first stack with Quick/Deep pipelines, SearXNG retrieval, and MCP tool-calls. No mandatory subscription. Your hardware, your models, your control.

Open Repository Live Demo Site

Live Demo is limited to 2 interactions. Install locally for full performance and unlimited runs.

No Mandatory Cloud Fee LM Studio Friendly Repo Grounding (`strict_repo_only`)

Quick Search for latency Deep Research for coverage MCP tools: quick / deep / fetch-url-context GitHub repo-scoped retrieval SearXNG JSON by default Dockerized local stack Quick Search for latency Deep Research for coverage

Why teams like it

Runs fully local on localhost
No forced paywall for every research run
Clear separation between quick and deep workflows

Quality control

Model choice drives depth and precision
Deep mode adds critic + synthesis quality gates
MCP can enforce repo scope for GitHub analysis

What you can show in a demo

Discovery → Deep research flow
Selection-to-Ask in Analysis view
MCP JSON-RPC calls with grounded context

Install In Under 2 Minutes

Use bootstrap for fastest setup, or manual commands if you prefer explicit control.

one-command setup

bash <(curl -fsSL https://raw.githubusercontent.com/zvspuentus-rgb/Argentic-Search-Lab/main/scripts/bootstrap.sh)

manual setup

git clone https://github.com/zvspuentus-rgb/Argentic-Search-Lab.git
cd Argentic-Search-Lab
cp .env.example .env
docker compose up -d --build

UIhttp://localhost:8093

MCPhttp://localhost:8193/mcp

SearXNGhttp://localhost:8393/search?q=test&format=json

Mode Comparison

Clear difference between speed-focused and quality-focused research paths.

Quick Search

Latency-first profile

Minimal orchestration
Fast answer turnaround
Great for iterative follow-ups

Deep Research

Coverage and confidence profile

Analyzer → Planner → Refiner
Multi-lane retrieval + critic
Cited synthesis + copilot follow-ups

Pipeline Visual

High-level execution map from query to grounded output.

MCP Tooling

Use JSON-RPC calls for quick/deep retrieval and URL context extraction.

Repo-grounded deep call

tools/call • search_deep

{
  "jsonrpc": "2.0",
  "id": 11,
  "method": "tools/call",
  "params": {
    "name": "search_deep",
    "arguments": {
      "query": "analyze this repo https://github.com/zvspuentus-rgb/Argentic-Search-Lab/tree/main",
      "limit": 10,
      "include_context": true,
      "context_max_urls": 8,
      "strict_repo_only": true
    }
  }
}

MCP flow

FAQ

Short answers to common setup and quality questions.

Is this really free?

Yes. Core usage is local-first with no mandatory subscription wall.

Why is Live Demo limited?

Hosted demo is quota-limited (2 interactions) to control free-tier resource usage. Local install is unrestricted.

How do I improve output quality?

Use stronger local models and Deep mode for high-stakes research tasks.

When should I enable strict_repo_only?

Whenever the query targets a specific GitHub repository and you want only repo-scoped evidence.