Symbiocene provider graph
We make environmental restoration legible to capital.
Symbiocene is building the graph of every environmental service provider, ranked for the buyers who must hire them. Each new job a buyer brings becomes a branch, so the next buyer with that job gets a ranked answer at once.
Watch the graph learnRequests from buyers appear here.
Click any point in the graph to see its condition, response, and place.
Drag or swipe the graph to turn it. Buyers, providers, counts, and timings are illustrative. The rules the requests cite are real and listed at the end of the page.
How the graph learns
- 1
A buyer names the job.
A data center operator, a sanitation agency, or a housing developer says what needs doing and where. The request enters at the root and moves from job to region to place, and at each step the first branch that fits takes it.
- 2
The model learns new jobs.
When a job is new to the graph, the model reads permits, contracts, and registries, ranks the providers it finds on past work, place, price, and results, and writes the answer in as a new branch.
- 3
Every buyer after that gets it at once.
The model tests the branch on the request that created it. From then on the graph answers that job in that place by itself, and the model’s time goes to whatever is new.
Each branch holds a condition, a response, and a place.
The condition says which requests the branch takes. The response ranks the providers for that job and place. The place sets the order the graph checks branches in.
The model writes branches as code that the team can read. This is the one it wrote for the first request in the demo.
{ condition: "place is Santa Clara County", response: "rank providers on past work, place, price and results", place: "under Water › Supply assessment › California, 1st of 1", learned: "request 1, a data center operator", rule: "Cal. AB 2469, in force Jan 1, 2027" }
New branches hang where they belong.
A directory puts every vendor on one long list, and each buyer searches it from the top. The graph files each new branch under the job and place it serves, so its shape becomes a map of who does which work, and where.
Sources and notes
Rules the requests cite
Method
The tree follows Sihan Xu’s essay How to Control an Agent with a Growing Behavior Tree (November 2026). In the essay, an agent’s reflex tree grows from what the agent lives through. Here, the tree routes buyer requests, and Symbiocene’s research agent plays the model.
What is illustrative
Buyers, providers, counts, distances, and timings are illustrative. The requests show the kinds of jobs that data centers, municipal sanitation agencies, and housing developers hire for.