Documentation

The colony,
documented.

How Stigmergy builds AI the way ants build cathedrals — from the bottom up, on Robinhood Chain. The full protocol, end to end.

On this page

Overview

Most AI is built inside walled gardens: a few companies own the models, data and compute, and everyone else rents access. The most important technology of the century is coordinated by command and owned by almost no one. Stigmergy is the bet that intelligence can be built like an ant colony — from millions of tiny local actions, each leaving a trace that others follow. It is a decentralized AI protocol on Robinhood Chain where anyone can contribute compute, data or validation; every contribution is recorded onchain, rewarded, and owned by whoever made it. No architect. No gatekeeper. Order, emergent.

What is stigmergy

In 1959 the biologist Pierre-Paul Grassé watched termites build without a blueprint. No leader, no plan, no direct communication — each insect simply leaves a trace (a pheromone mark) and reacts to the traces left by others. From this, structure emerges that no single insect could conceive. He named the mechanism stigmergy: coordination through the environment itself. Stigmergy the protocol applies it to machine intelligence — the “environment” is the chain, the “traces” are onchain tasks, results and validations, and the emergent structure is an open AI network.

Architecture

Three contracts, no admin over user funds.

StigToken

The $STIG ERC-20. Fixed 1B supply, minted once. Staked to run nodes and spent as task fees.

StigmergyNodes

The staking & reward core. Holds each operator's stake and streams rewards stake-weighted.

TaskRegistry

The workload ledger. Escrows task fees and routes them to the node pool on validation.

The registry never holds rewards — on validation it approves and pushes the fee straight into StigmergyNodes via notifyReward. The token is inert infrastructure: it cannot be minted, paused or frozen.

Node lifecycle

Task lifecycle

Each task walks four onchain states — the trace it leaves behind.

Open  submitted with a fee, awaiting a nodesubmitTask
Claimed  an active node took the taskclaimTask
Completed  the worker posted a result hashcompleteTask
Validated  a second node confirmed it — fee routedvalidateTask

The validator must be a different node than the worker — no one signs off on their own work. On validation the fee leaves escrow and enters the reward pool.

Rewards & math

Node rewards come from three traces: protocol fees from every validated task, token emissions that bootstrap growth, and a dedicated 3% of total supply reserved for node & staking rewards. All three feed one pool through a single stake-weighted accumulator — the MasterChef pattern — so rewards split pro-rata to stake, gas-cheap and fair no matter how many nodes join.

1. The accumulator. Let S = accRewardPerShare (scaled by 1e18) and T = totalStaked. Every reward R pushed via notifyReward(R) updates:

St+1  =  St  +  R · 1e18 / T

2. Pending rewards. For a node i with stake sᵢ and stored debt dᵢ:

pendingi  =  si · S / 1e18  −  di

3. A node's slice of one reward. Because S rises by R·1e18/T, node i's share of that reward R is exactly its stake weight:

rewardi  =  R · si / T   =   R · (stake share of node i)

4. Debt reset. On any stake change (register / addStake / unstake / claim), the node's debt is re-based to the current accumulator so it only ever earns on rewards added while it was staked:

di  ←  si · S / 1e18

5. Yield over a period. A node's harvest between two accumulator snapshots S₀ → S₁:

Δrewardi  =  si · (S1 − S0) / 1e18

6. Staking APR. With annual inflow Ryear = fees + emissions + allocation streamed over a year, the pool-wide APR (before compounding) is simply the inflow over the staked base:

APR  ≈  Ryear / T   ×  100%

Every node earns the same rate on its stake; a larger stake earns proportionally more, never a better rate. As T grows, per-token APR falls — the colony self-balances.

7. Emissions schedule. Emissions bootstrap the early network and taper as real fees take over. A simple per-epoch geometric decay with factor ρ ∈ (0,1):

Et  =  E0 · ρt     ⟹   Σt=0 Et  =  E0 / (1 − ρ)

The total emitted is bounded, so emissions can never exceed the budget set aside for them — the tail is finite by construction.

8. Fee routing. When a task is validated, its escrowed fee f leaves the registry and enters the pool, so R = f in equations 1–3 — real AI usage flows straight to the operators securing it:

on validate:   T unchanged,   S  +=  f · 1e18 / T

Tokenomics

Total supply1,000,000,000 STIG (fixed)
Node & staking allocation3% · 30,000,000 STIG
Deployer / launch / protocol97% · 970,000,000 STIG
Mint functionnone — minted once at genesis
Transfer tax / fee0%

$STIG is the coordination asset: staked to secure the network and spent as the fee that summons the colony to a task. Its supply is fixed and its contract is ownerless.

Trust model

FAQ

Do I need special hardware to run a node?You need enough $STIG to meet minStake and infrastructure able to execute the workloads you claim. The endpoint you publish is where work is served.
When are rewards paid?Continuously — they accrue every time notifyReward is called (on each validated task, plus emissions/allocation). You harvest them with claim().
Can I lose my stake?The core contract never confiscates stake; unstake returns it in full. Slashing for bad work is a governance extension, not present in the base contracts.
What chain is this on?Robinhood Chain (id 4663).

Glossary

TraceAny onchain contribution — a task, a result, a validation — that others build on.
NodeA staked operator that executes and validates AI workloads.
minStakeThe minimum $STIG required to be an active node.
notifyRewardThe call that adds $STIG to the reward pool and streams it to stakers.
Worker / ValidatorThe node that executed a task vs. the different node that confirmed it.
← Back to home Follow @stigmergy_ai 🐜