Research Agents
There is no single AI in EcoDesign. Each stage of a project has its own family of specialist agents, and they hand work to each other in a fixed order. This page says who they are and where you meet them.
1. Agents belong to stages, not to you
You do not pick an assistant and stay with it. You open a step, and the agents that step needs run for it. The app currently runs around eighty-five of them across the four stages — Research, Design, Application & Construction and Farm — and most of them you never address directly.
The Project Map is where agents are launched
Every agent-backed step is a node on your Project Map. Click a node and the panel tells you what that step needs before it can run, what it produced last time, and gives you the button that starts it. Working from the map is how you see the whole family at once.
2. The research advisors
Six advisors read your site data and answer in conversation. Each one is a pair: an agent that gives the first, long assessment, and a second that handles your follow-up questions on it. That is why a follow-up costs less than the original — it is a different, cheaper exchange.
| Advisor | Where you meet it | What it covers |
|---|---|---|
| TerraScope Advisor | TerraScope Research | Overall site conditions, vulnerabilities, regenerative actions |
| Soil Advisor | Soil Insights | Soil improvement, nutrients, amendments, management |
| Climate-Smart Advisor | Climate-Smart EcoDesign | Climate resilience, sustainable building, adaptation |
| Water Advisor | Water research | Water sources, storage, irrigation and management |
| EcoGarden Advisor | EcoGarden Insights | Garden design and permaculture advice |
| Fire Resilience Analyst | Fire Resilience | Wildfire risk and fire-resistant design |
A seventh, the Resilience Game Analyst, scores the systems and practices you grouped into zones and tells you where they reinforce each other and where they do not.
Environmental Research, Local Demographics and Indigenous Knowledge are not advisors. They search current web sources and return a cited report, so they take longer and they answer once rather than in conversation.
3. The design families
Design is where the agents stop being conversational and start being pipelines. You give a step its inputs, the pipeline runs, and you accept or reject what comes out.
Element selection
Flora, fauna, structures and technologies each get the same three-agent treatment: a pre-filter narrows the catalogue to what could plausibly work on your site, a selector proposes a set with reasons, and a suitability reviewer checks the proposal back against your conditions. Soil and substrate selection works the same way.
Site studio
The Sector Analysis Advisor reads your drawn site and reports on sun, wind, water and access sectors. The Zone Component Suggester proposes components for a specific zone once you have drawn your permaculture zones.
Plant Guild Designer
Nine agents in sequence compose an ecologically optimised guild from the flora you selected: a pre-filter, a context analyser, a role mapper, a layer architect, then synergy, spatial and temporal optimisers, a composer that assembles the result, and a visualiser. A tenth, the Companion Suggester, helps you build the plant set that feeds the pipeline.
Passive solar design
Eight solar agents produce the building assessment — site solar, orientation and layout, glazing, thermal mass, shading, envelope and ventilation, heating and cooling, and an integration report. What follows them is deliberately not an agent: the floor plan geometry itself is generated algorithmically, because that is the part language models get wrong.
Practice transition
Nine agents assess moving an existing production system to sustainable practice — current state, site capability, pest and disease risk, systems recommendation, practice selection, compatibility, risk and timeline, an IPM strategy, and a feasibility report.
4. The farm agents
The Farm stage runs six AI-guided problem domains — pests, weather, irrigation, soil, transition and crop planning — on a Farmer Hub subscription for that project. Farm responses are metered at twice the research rate.
5. Two agents that work on every project
The Project Guide
The assistant in the Project Map that knows your project's state and answers "what should I do next?". It reads your stages, your findings and what is locked, so its answers are about your project rather than the product in general. It charges a flat 2 credits per turn.
The findings extractor
Runs after every research report is saved and distils it into the structured facts later steps read. You never address it and it never charges you.
6. Working with them well
Fill in the site data first
Every advisor reads your project's climate, soil, water and vegetation tables before it answers. An empty table does not stop the agent — it makes it generic. The Research Hub shows you which tables are still empty.
Order matters more than volume
Agents downstream read the findings of agents upstream. Running the guild pipeline before you have selected flora, or the floor plan before the solar assessment, gets you an answer built on nothing. The Project Map greys out a step until its inputs exist for exactly this reason.
7. Where to go next
AI research overview
The research stage end to end — phases, findings, staleness and cost.
Your research library
Everything the agents have told you, filtered and searchable.
Design tools
The design families in detail, and how to unlock them.
Credits
What each kind of agent response costs and how credits are topped up.