A Name from Mycology
When we were looking for a name, we didn't want another acronym, another portmanteau of "AI" and something, or another Latin word that sounds impressive but means nothing specific. We wanted a name that actually described the architecture we were building.
We found it in mycology.
A thallus is the entire vegetative body of a fungal organism — the undifferentiated structure that includes both the vast underground mycelial network and the visible fruiting bodies that emerge from it. It's not the mushroom you see above ground. It's not just the hidden network of threads beneath the soil. It's the whole organism, working as one.
That's exactly what an AI agent platform should be.
The Mycelial Network
Beneath the forest floor, mycelial networks span enormous distances. A single network can connect thousands of trees, moving nutrients, water, and chemical signals between them. Mycologists call this the "Wood Wide Web" — a decentralized intelligence system where no single node controls the network, but the network as a whole solves problems that no individual node could solve alone.
This is the architecture we built for agent-to-agent communication.
When Thallus processes a complex question — say, "How does our vendor spend trend compare to contract renewal terms across regions?" — it doesn't route that to a single monolithic model. It decomposes the question into a plan, assigns specialized agents to each step, and lets them share context through a common board. One agent profiles the database schema. Another queries the ERP. A third searches contract documents. A fourth synthesizes their findings.
Each agent is a node in the network. The board is the mycelium — the shared substrate through which information flows. No single agent sees the whole picture. The network does.
This isn't just a metaphor. Mycelial networks exhibit properties that map directly to how we designed the system:
Decentralized coordination. There's no master node in a mycelial network. Nutrients flow toward where they're needed based on local chemical gradients. In Thallus, the planner creates a dependency graph but doesn't micromanage execution. Agents run when their dependencies are met, share results on the board, and the system adapts.
Adaptive growth. Mycelial networks grow toward nutrient sources and away from threats. When the network encounters a new food source, it extends. When a pathway is blocked, it reroutes. In Thallus, the evaluation loop after each execution batch decides whether the plan is complete, needs additional steps, or should be restructured entirely. The system grows its execution plan based on what it discovers.
Resource sharing across species. Mycelial networks don't just connect trees of the same species — they connect different species, enabling cross-species nutrient transfer. In Thallus, agents connect different data sources — SQL databases, document stores, APIs, spreadsheets — and share findings across those boundaries. A database query result informs a document search. A document finding refines a follow-up query. The value emerges from the connections between different systems.
The Fruiting Body
Most people think of mushrooms as the organism. They're not. The mushroom is the fruiting body — the visible, temporary structure that the mycelial network produces when conditions are right. It's the network's way of making itself useful to the world above ground.
In Thallus, the fruiting body is the response you see. The synthesized answer. The chart. The cited analysis. The workflow that executes automatically.
All the complexity — the planning, the parallel agent execution, the cross-source reasoning, the governance checks — happens underground. What surfaces is clean, cited, and actionable. You asked a question. You got an answer that draws from six different systems, with every claim traceable to its source.
The fruiting body isn't the intelligence. It's the expression of the intelligence. The intelligence is the network.
Emergent Intelligence
The most fascinating property of mycelial networks is emergence. Individual hyphal threads follow simple rules — grow toward nutrients, avoid toxins, branch when conditions allow. But the network as a whole exhibits complex problem-solving behavior that no individual thread was programmed to perform.
Slime molds (which share the thallus body plan) have been shown to solve maze problems, replicate efficient transport networks, and optimize resource allocation — all without a central brain, all from simple local rules applied across a connected network.
This is what happens when you connect specialized AI agents through a shared context space. The data explorer doesn't know how to search documents. The document analyst doesn't know how to write SQL. The synthesizer doesn't know how to query APIs. But connected through the board, with a planner that understands dependencies, the system answers questions that no individual agent could answer alone.
The intelligence is emergent. It comes from the connections, not the components.
Why It Matters
We could have named the platform something generic. But "Thallus" carries a specific claim: that the value of an AI agent platform isn't in any single model or any single capability. It's in the network — the connections between agents, between data sources, between the questions you ask and the systems that hold the answers.
A better model makes each agent smarter. A better network makes the whole system smarter. We're building the network.
That's why we named it Thallus.