An evidence-first autonomous agent ecosystem for economic intelligence, prediction markets, Web3 research, and bounded machine action.
Public experience · Quick start · Operator guide · Architecture · Economic Habitat · Brand · Meridian
NOEMA is built around a simple rule:
Increase autonomy only when evidence, economics, and authority boundaries can support it.
The system observes markets and other public information, records forecasts and research state, evaluates later outcomes chronologically, allocates bounded research attention across specialists, and exposes the resulting system as an auditable operating world.
The long-term direction is broader than a trading bot: NOEMA is an agent economy and intelligence runtime that can turn observed information into structured research, bounded decisions, and eventually real-world action — without hiding uncertainty or silently expanding its own authority.
NOEMA separates implemented capability from future authority.
| Layer | Current repository state |
|---|---|
| Research runtime | Persistent specialist ecosystem, bounded research attention, missions, experiments, evidence, immutable forecasts, and outcome evaluation |
| Cognition | Optional model-assisted reasoning behind explicit configuration and budget gates |
| Prediction markets | Kalshi and Polymarket integrations for market/account observation where configured; paper research remains the default operating mode |
| Web3 | Trench research path for on-chain discovery and forward evidence collection |
| Meridian | Frozen request/review bridge for world-intelligence context; no always-on shared service is implied |
| Economic OS | Cash, reservations, costs, budgets, measured outcomes, and reconciliation rules remain distinct |
| Wallets | Observed wallet/network state can be surfaced; visibility never grants signing or transfer authority |
| Execution | Deterministic gateway exists and is fail-closed by default; a visible gateway is not proof of live trading |
| Ops Console | Runtime-backed Economic Habitat, inspectors, replay, live capital surfaces, work packets, and Autonomous Desk |
| Public site | Static documentation/report explorer; it is not the private worker or authority surface |
Paper results are not revenue. Wallet observations are not authority. Missing state stays missing.
See Economic integrity and Production boundaries.
NOEMA's console is no longer just a table of records. It projects canonical runtime state into a spatial operating world.
The habitat organizes real entities into functional zones:
- Cognition Array — model and data providers
- Agent Colony — registered specialists and handoff participants
- Mission Control — persisted active missions
- Market Deck — markets and immutable forecasts
- Evidence Vault — evidence, sessions, experiments, lessons, and recorded outputs
- Treasury & Gateway — wallets, runtime tools, policy, and execution boundary
Agents move or light rooms only when persisted assignments, handoffs, or mission relationships support the visualization. Historical replay hides later/current-only state.
Phase 3 adds an auditable, single-responsibility desk model:
CHIEF → SCOUT → MAP / CONTEXT → VET → ODDS → SIZE → EXECUTION → RISK / EXIT
These are presentation roles over canonical entities, not magically-created agents. If NOEMA cannot defensibly bind a real registered specialist or tool to a seat, the seat stays:
VACANT / UNBOUND
Active Work Packets reconstruct mission routes from exact assignments and persisted handoffs. The Shift Tape reuses NOEMA's canonical event timeline.
Read the full contract in Economic Habitat.
flowchart LR
A[Market / Web3 / World Sources] --> B[Perception + Research]
M[Meridian Context] --> B
T[Trench On-chain Research] --> B
B --> C[Specialist Ecosystem]
C --> D[Autonomous Desk]
D --> E[Evidence + Immutable Forecasts]
E --> F[Economics + Deterministic Risk]
F -->|reject| P[PASS / Kill]
F -->|paper-authorized| Q[Paper Action]
F -->|all live gates satisfied| G[Execution Gateway]
Q --> H[Outcomes + Scoring]
G --> H
H --> I[Calibration + Specialist Evolution]
I --> C
The graph describes flow, not current live authority. The execution gateway remains separately gated and fail-closed.
Evidence over theater. The UI is allowed to look alive only when canonical state supports it.
Bounded specialists. Research attention is allocated across specialist families instead of assuming one universal model should do everything.
Chronological evaluation. Forecasts are recorded before outcomes and scored later; winning streaks are not promotion criteria.
Economic honesty. Model/API/compute costs, cash, reservations, paper outcomes, and reconciled economic value remain separate concepts.
Deterministic authority. Model output can propose; deterministic policy governs whether action is even eligible.
Restraint is valid. NOEMA may leave capital, compute, or attention unused when nothing clears the evidence threshold.
| Phase | State | Direction |
|---|---|---|
| 01 · Reality-first console | ✅ Implemented | Evidence-backed operations view, replay, diagnostics, economics |
| 02 · Living habitat | ✅ Implemented | Inhabitants, workstations, mission occupancy, handoffs, recorded outputs |
| 03 · Autonomous Desk | ✅ In progress | Single-responsibility seats, active work packets, shift tape |
| 03B · Rule Rack / Kill Board / Shift Report | ⏭ Next | Surface governing rules, rejected opportunities and exact rejection reasons, end-of-shift audit |
| 04 · Policy-gated execution | 🔒 Gated | Only after explicit authority, supported adapters, budgets, evidence, and reconciliation |
| 05 · Real-world autonomous impact | ◌ Vision | Expand from research loops into legitimate external work with the same audit discipline |
The roadmap is directional. It does not grant execution authority or claim profitability.
Python 3.12+, from a checkout:
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
noema agent-once
noema doctor
noema ladderRun the continuous paper worker:
noema-agentRun the private/local Ops Console:
noema-dashboard
# http://127.0.0.1:8787Public observation paths may require network access. Optional providers, authenticated account observation, cognition, and hosted services require their own configuration. See Quick connect.
noema economics-report > economics-report.jsonThe public explorer can import that report locally in the browser. The importer checks consistency; it does not authenticate a runtime or prove receipts.
Full net economic profit remains unknown until the relevant costs, transfers, balances, and realized outcomes are reconciled.
ruff check .
pytest -q
npm ci
npm run site:test
npm run ops:test
npm run site:build
npx playwright install chromium
npm run site:qa
npm run ops:qaThe repository's NOEMA Verify workflow runs the Python suite, Node contracts, static-site build, and production browser QA on changes.
NOEMA and Meridian have separate responsibilities.
Meridian organizes world evidence, entities, relationships, events, and timelines.
NOEMA consumes frozen research context, performs bounded economic research, records forecasts/decisions, evaluates outcomes, and enforces its own authority/economic boundaries.
The bridge is intentionally explicit and hash-bound. Meridian context does not become an order merely because NOEMA can read it.
See Meridian ↔ NOEMA contract.
| Area | Purpose |
|---|---|
noema/ |
Python runtime, research, economics, venue/wallet observation, mission and persistence logic |
noema/static/ |
Private Ops Console and runtime-backed habitat UI |
site/ |
Static public experience and local report explorer |
docs/ |
Architecture, operating contracts, evidence/economic boundaries, setup |
tests/ |
Python + Node regression and truth-contract tests |
scripts/ |
Browser QA and build/verification tooling |
render.yaml |
Hosted deployment blueprint/configuration; configuration is not proof of current deployed health |
- Master mission
- Economic Habitat + Autonomous Desk
- Agent ecosystem
- Live workstation
- Economic integrity
- Prediction venues
- Trench-1
- Agent wallet
- Production boundaries
- Brand system
Evidence → Belief → Edge → Restraint
NOEMA is an experimental research and agent-runtime project. Nothing in this repository is a guarantee of profit, a promise of self-funding, or automatic authorization to place orders or move funds.