PatrickFanella b311fce7ea
CodeQL / Analyze Go (push) Skipped
CodeQL / Analyze TypeScript/JavaScript (push) Skipped
Security / Frontend Dependency Audit (push) Successful in 12s
CD / Publish Docker image (push) Failing after 13s
Security / Go Dependency Audit (push) Successful in 37s
Security / Docker Image Scan (push) Skipped
CI / Unit tests (push) Successful in 3m20s
CI / Lint (push) Successful in 3m40s
CI / Detect backend changes (push) Successful in 8s
Security / Secret Detection (push) Successful in 30s
CI / Integration tests (push) Successful in 33m1s
CI / Smoke tests (push) Successful in 1m33s
CI / Build (push) Successful in 59s
Merge pull request 'Unblock paper trading across venues and surface every silent gate' (#52) from t3code/review-trading-system-backend into main
Reviewed-on: #52
2026-09-23 20:27:30 -05:00
2026-08-15 00:13:29 -05:00
2026-08-21 15:31:07 -05:00
2026-09-21 10:23:29 -05:00

Augr

Augr is a paper-first, multi-market trading research and execution system. A Go service schedules data collection and strategy pipelines, applies hard risk controls, persists evidence in PostgreSQL, and exposes a React operator UI.

Augr can submit orders when a deployment is deliberately configured for it, but live trading is disabled by default. A healthy process or a successful scan does not by itself prove that a strategy is qualified or that an order was sent.

What is included

  • Scheduled and manual strategy pipelines for stocks, options, crypto, Kalshi, and Polymarket-oriented workflows.
  • Market-data adapters for Polygon, Alpha Vantage, Finnhub, Financial Modeling Prep, Yahoo, Alpaca, Tradier, Binance, Kalshi, and Polymarket.
  • Paper execution plus guarded broker adapters for Alpaca, Binance, Kalshi, and Polymarket.
  • Persistent runs, decisions, events, orders, trades, positions, accounting evidence, provider observations, and audit records.
  • Kill switches, circuit breakers, exposure limits, reconciliation, projection checks, and capability-scoped readiness reporting.
  • JWT and API-key authentication, WebSocket activity streaming, Prometheus metrics, and Grafana dashboards.

Repository map

Path Purpose
cmd/tradingagent/ CLI, application bootstrap, scheduler wiring, and strategy runners
internal/ API, domain, provider, execution, risk, automation, and persistence code
migrations/ Ordered PostgreSQL/TimescaleDB migrations
web/ React, TypeScript, and Vite operator UI
monitoring/ Prometheus rules and Grafana provisioning
scripts/ Maintained release, recovery, verification, and operator helpers
docs/ Canonical guides, ADRs, and current runbooks

Quick start

Prerequisites are Go 1.25.13, Node.js 22, npm, Python 3, Docker Compose, Task, and the migrate CLI.

git clone https://git.subcult.tv/subculture-collective/augr.git
cd augr
cp .env.example .env

For local development, set at least:

POSTGRES_PASSWORD=postgres
DATABASE_URL=postgres://postgres:postgres@localhost:5434/tradingagent?sslmode=disable
REDIS_URL=redis://localhost:6380/0
JWT_SECRET=replace-with-a-long-random-value
PROJECTION_ACCOUNT_ID=00000000-0000-4000-8000-000000000064

# Configure one supported LLM provider and one primary data provider.
OPENAI_API_KEY=...
POLYGON_API_KEY=...

Then start the dependencies, migrate, and run the applications:

docker compose up -d postgres redis
task migrate:up
task build
./bin/tradingagent serve

In another terminal:

npm --prefix web ci
npm --prefix web run dev

The native API is available at http://localhost:8080; Vite uses http://localhost:5173. The Compose-published database and cache ports are 5434 and 6380. Running the application itself through Compose publishes it at http://localhost:8081.

Fresh disposable databases seed a local account named [email protected]. The retained development password is demo-pass; never reuse that credential in a shared or deployed environment.

See Getting Started for the complete local flow and Development Setup for configuration and validation.

Common commands

task build             # compile ./bin/tradingagent
task test              # short Go suite
task test:race         # short Go suite with the race detector
task test:maintenance  # verify the integration harness itself
task web:check         # frontend lint, tests, and production build
task test:integration  # full contracts against TEST_DATABASE_URL
task audit             # Go vet, lint, vulnerability, and format checks

The database integration task is intentionally fail-closed: use a migrated, disposable database that is not the development or production database. See Testing for the test tiers and their ownership.

CLI

tradingagent provides these top-level commands:

  • serve starts the API and optional scheduler.
  • run executes the configured one-shot workflow.
  • strategies manages strategies and manual runs through the API.
  • automation inspects or operates automation jobs.
  • portfolio, risk, and capital-ladder expose operator readbacks and controls.
  • memories manages agent memories.
  • dashboard starts the terminal dashboard.

Run ./bin/tradingagent --help and <command> --help for the current flags.

Safety and operational scope

  • ENABLE_LIVE_TRADING=false is the default and should remain so until the release and venue-specific gates are satisfied.
  • Scheduled jobs are independently capable of succeeding, degrading, failing, or being disabled. Container health is not workflow health.
  • Schema-affecting releases are applied migration-first, followed by the app, then exact schema and health readback.
  • Qualification evidence belongs in immutable external ledgers and release artifacts, not generated files committed to this repository.

Operators should start with the Runbooks. Current limitations are tracked in Known Issues.

Documentation

License

See LICENSE.

https://www2.onnwee.me

S
Description
Autonomous multi-agent trading system in Go with LLM-powered analysis, debate, and execution pipelines
Readme GPL-3.0
79 MiB
Languages
Go 88.6%
PLpgSQL 4.9%
TypeScript 4.6%
Python 0.7%
Shell 0.6%
Other 0.5%