The $1 Property Report: Turning DeepSeek Harness Into a Senior Zoning Researcher

Realtors, developers, and architects — the pre-construction research department you couldn't afford is now an open-source AI preset. Here is exactly what it does, how the multi-agent swarm works, what it costs, and why it refuses to guess.
Classification: Product Launch | Status: Open-Source Release | Published: August 2026

In the 1970s, hip-hop pioneers took cheap turntables and scratched-up vinyl, breaking the linear 33⅓ RPM playback to turn a consumer machine into an instrument. They took records nobody wanted and rearranged them by refusing to use the manual.

We did the exact same thing to an AI coding platform.

DeepSeek Harness (DSH) is designed as a developer tool: a framework with shell access, subagent orchestration, and a model router. Instead of building another coding assistant, AxoWorks rewired it. The result is the Property & Code Researcher: an open-source, multi-agent preset that acts as an obsessive senior property researcher with the work ethic of someone who reads footnotes for fun. It reads municipal codes, parses zoning maps, and grades its own evidence, all for the price of a used GPU hour.

1. Two Tactical Modes: Deep Research vs. Gut Check

Give the agent an address and ask hard feasibility questions. You control the deployment:

• Gut Check Mode: A rapid, single-model pass. It uses a strict call budget to deliver a verdict-first snapshot of a parcel's viability in minutes.
• Deep Research Mode: The full asynchronous swarm. It fans out parallel harvesters to scrape GIS data and permit histories, triggering a Red Team vs. Blue Team adversarial debate over the findings, overseen by a Neutral Arbiter.

2. What It Actually Does (The Zero-Hallucination Protocol)

Ask it: “Can the client enclose the side porch under the bedroom into a music room?” It does not guess from a generic LLM search index. It answers strictly from local records.

• Definitive Zoning Resolution: It reads the governing city or county GIS for the ordinance number and effective date. If a printed county assessor card is stale but the city's live GIS states otherwise, it defaults to the governing city record.
• Verbatim Dimensional Standards: Setbacks, height limits, and lot coverage are extracted and quoted verbatim from the operative code chapter, with exact section citations attached.
• Critical Area Overlays: It builds a hit-or-empty matrix for geohazards, flood zones, and wildfire timing. An empty query result is actively reported as a verified clean finding, never silently skipped.
• The Deadlock Section: If two municipal records contradict each other, the AI does not synthesize a hallucinated compromise. It logs a “Deadlock” and tells you exactly which municipal department to call to resolve the discrepancy.

Every claim in the final report carries a confidence grade: [Confirmed], [Confirmed-as-printed], or ⚠ UNVERIFIED(reason).

3. The Architecture: Why DeepSeek Harness?

We didn't build a basic chatbot wrapper. We leveraged DSH’s underlying Cordis microkernel because it treats everything as a hot-swappable plugin.

• Asynchronous Harvester Fan-Out: The workflow engine allows the agent to spawn parallel sub-tasks. GIS data runs in one channel, permit history in another, and cost data in a third. They all collide back into the arbiter.
• Dynamic Model Routing: Real estate research is bimodal (90% data retrieval, 10% legal reasoning). DSH lets us route the heavy reasoning to frontier models and the raw data parsing to free, local open-weight models (like Qwen via Ollama) to keep the cost under $1 per report.
• Built for Hostile Infrastructure: Municipal websites are notoriously difficult to scrape. When our early PowerShell extraction scripts triggered heuristic Antivirus alarms (flagged for fileless in-memory decompression while aggressively bypassing Cloudflare), we used DSH’s plugin architecture to swap in robust, native HTTP response handlers and dedicated browser bridges. It stops for CAPTCHAs, refuses to bypass TLS security without logging it, and handles 403 blocks gracefully.

4. The Hustle: Zero-Click Installation

The entire preset is MIT-licensed and free. You only pay for the API calls (or run it locally for $0).

git clone https://github.com/Axotopia/dsh-property-researcher
cd dsh-property-researcher
INSTALL.cmd
      

Restart DeepSeek Harness, select the property-researcher preset, and type:

Gut Check 1200 Main St, Vancouver WA
      

INSTALL.cmd copies the preset into place and installs the bundled browser tool dependencies. There is nothing else to configure.

Frequently Asked Questions (FAQ)

What is the AxoWorks Property Researcher?
The AxoWorks Property Researcher is an open-source (MIT-licensed) multi-agent AI preset for DeepSeek Harness. It automates real estate due diligence — including zoning code extraction, site feasibility, and permit mapping — while enforcing verbatim code citations and evidence grading for every claim.

How does the Property Researcher eliminate AI hallucinations?
The system uses a strict Citation Protocol and an adversarial architecture. Every claim must trace back to a municipal record fetched during the session, with a confidence grade attached. Unverifiable claims are explicitly flagged as ⚠ UNVERIFIED, and conflicting data triggers a “Deadlock” section that identifies the exact phone call needed to resolve the gap, rather than generating a hallucinated compromise.

What is DeepSeek Harness (DSH)?
DeepSeek Harness is an installable open-source agent platform. It provides AI models with developer-grade tools, including file system access, subagent orchestration, workflow automation, and dynamic model routing, allowing users to deploy complex, modular workflows.

How much does the AxoWorks Property Researcher cost to run?
The software preset itself is 100% free. Users only pay the underlying model API usage costs. A comprehensive “Deep Research” multi-agent report typically costs under $1.00 via API, while a fast “Gut Check” screen costs only a few cents. Costs scale based on the specific models routed by the user.

What jurisdictions does the Property Researcher cover?
The preset is field-tested on Washington State public records, including Seattle, King County, and Chelan County across both incorporated cities and unincorporated systems. For unmapped jurisdictions, it executes an automatic discovery ladder to locate public GIS and code endpoints; untested territory is labeled as such in the final report.

Can the tool bypass paywalls or government login portals?
No. The Property Researcher strictly reads public records. It renders JavaScript-walled public pages through a dedicated browser bridge, but it treats CAPTCHAs and challenge pages as terminal results, stopping for human verification rather than brute-forcing them. It never submits forms or credentials unasked.

Does this AI replace architects, surveyors, or land use attorneys?
No. The Property Researcher acts as a preliminary research department to organize municipal data, highlight critical area blocks, and prepare feasibility packets. It explicitly states in every report that it is not a substitute for stamped engineering, professional surveys, or official legal advice.