echolon

See what others can't.

echolon is the library that takes a certified trading strategy from a release bundle to a broker order. It is deliberately small.

panel/    the market-data snapshot format (bars, contracts, sector, QC)
score/    the SignalEngine contract every strategy implements
bundle/   hash-verified release manifests (build, verify, load)
live/     bundle runtime -> target lots -> MiniQMT orders, plus the paper
          fill model, the book risk overlay and the position reconciler

That is the whole product. 41 Python files.

What it is not

echolon is not a backtest engine, an optimiser, an indicator library, or an agent toolkit. It used to advertise all four. On 2026-08-18 the old framework — the Backtrader engine wrapper, the Optuna study, the 214-indicator TA-Lib catalog, the five-component strategy system, the MCP server, the 23 in-package agent skills, the data extractors, the per-exchange market adapters, the slot-era live tree and the M1 book programme — was retired in one commit, because nothing in the current framework reached any of it.

If you need that code it is not gone, it is elsewhere: tag checkpoint-2026-08-18-pre-echolon-retirement, branch legacy/pre-retirement-2026-08-18. See ARCHITECTURE.md for what remains and why, and CHANGELOG.md 0.4.0 for the full removal list.

Breaking: echolon and echolon-mcp are no longer installed. echolon hello, echolon init, echolon backtest, claude mcp add … echolon-mcp, echolon.run_backtest(...) and every from echolon import BacktestConfig-style re-export are gone. There is no compatibility shim; this is intentional.

Related MCP server: BotSpot

Install

pip install echolon

Five runtime dependencies: pandas, numpy, pydantic, pyarrow, structlog. MiniQMT execution additionally needs xtquant, which is Windows-only and ships with the broker terminal rather than from PyPI — the rest of the library works without it.

Use

Load a hash-verified bundle into a runnable strategy:

from echolon.live.book.bundle_runtime import load_bundle_strategy

runtime = load_bundle_strategy("bundles/composite_daily_v1_prod_1")
runtime.strategy      # SmokeStrategy, built from the manifest's constructor
runtime.engines       # the SignalEngine instances the manifest pins
runtime.rebalance_rule

load_bundle_strategy recomputes every file hash before importing anything, requires each signal module to be self-contained and to define exactly one SignalEngine, checks the loaded engine's signal_id/family against the manifest, and refuses a constructor it cannot build. Every one of those is a hard failure with no override.

Build and verify bundles from the command line:

echolon-bundle --help     # the only console script

Write a signal:

from echolon.score import ScoreVector, SignalEngine

class MySignal(SignalEngine):
    signal_id = "my_signal_v1"
    family = "tsmom"

    def compute(self, view) -> ScoreVector:
        ...

Status

0.4.0. Live MiniQMT execution is the part that is shipped and used; it is what the rest of the library exists to serve. Bring your own market data — echolon defines the panel format and reads it, but no longer extracts it.

Development

pip install -e ".[dev]"
pytest tests/ -q -ra

-ra is not cosmetic: a public checkout has no artifact store, so the store-backed falsifiers skip, and a skip is indistinguishable from a pass in a bare summary.

Licence: Apache-2.0.