🧿 backtest-ollama-crontab

A TypeScript monorepo for backtest-kit demonstrating how a local Ollama LLM (running gpt-oss quantized) can be wired into a trading-signal pipeline as an outline-based risk filter, with a 15-minute crontab doing live Telegram-channel ingestion. The strategy parses signals from any public Telegram channel, asks the LLM whether each signal is safe to trade

📚 API Reference | 🌟 Quick Start

🚀 Quick Start

1. Build the workspace packages

npm run build:x        # macOS / Linux
npm run build:win      # Windows

Each build emits per-package build/index.cjs + rolled-up types.d.ts.

2. Authenticate the Telegram MTProto session

cd ./packages/main
npm run auth

A QR code prints in the terminal. Scan it with the Telegram app on your phone (Settings → Devices → Link Desktop Device). On success, a session.txt file is written next to the auth command.

3. Copy the session into the strategy folder

cp packages/main/session.txt content/jan_2026.strategy/session.txt

The strategy reads this file at runtime to subscribe to the configured channel without re-authenticating.

4. Run the backtest

From the repo root:

npm start -- --backtest --ui --entry ./content/jan_2026.strategy/jan_2026.strategy.ts

🎯 What this repo demonstrates

A real-world experiment: take any free, public Telegram trading-signals channel, wire each signal through a local Ollama LLM (gpt-oss quantized) acting as a risk filter, and measure whether the LLM gate actually improves P&L on historical data.

Pipeline

Telegram crawler (packages/core/src/lib/services/core/CrawlerService.ts) — pulls raw messages from the configured channel.
Parser (packages/core/src/lib/services/screen/ChannelScreenService.ts) — extracts direction, entry, targets, stoploss from signal text into a parser-items Mongo collection.
Risk outline (packages/core/src/logic/outline/risk.outline.ts) — for every parsed signal, runs the LLM against 1m/15m candles + a pre-computed metrics packet (avgRangePct, momentum24hPct) and asks for a riskAction: "skip" | "follow" verdict.
Strategy (content/jan_2026.strategy/jan_2026.strategy.ts) — opens a position only when the LLM voted follow and the live price falls inside the channel's entry zone. TP = targets[2], SL = signal.stoploss.
Crontab triggers — Cron.register from backtest-kit schedules a 15m live-fetch and a one-shot backtest-prepare handler. In live mode the worker re-polls the channel every 15 minutes; in backtest mode it pulls the entire frame at startup.

📊 Before vs After Ollama

Two backtests on the same parsed-signal set for January 2026:

📋 Trade Log WITHOUT Ollama

Signals

#	Symbol	Opened (UTC)	Closed (UTC)	Held	Dir	Open	Close	Peak	DD	PNL%	Exit
1	SOL	Jan 05 07:13	Jan 05 20:09	12.9h	LONG	$135.23	$139.10	+2.42%	−1.72%	+2.45%	take-profit
2	NEAR	Jan 06 08:46	Jan 06 11:44	3.0h	LONG	$1.77	$1.80	+1.57%	−0.59%	+1.64%	take-profit
3	HYPE	Jan 06 09:59	Jan 06 12:13	2.2h	LONG	$26.52	$27.04	+1.55%	−1.41%	+1.56%	take-profit
4	TRX	Jan 06 10:17	Jan 11 00:06	109.8h	SHORT	$0.2915	$0.3027	−0.32%	−4.21%	−4.24%	stop-loss
5	FARTCOIN	Jan 06 10:43	Jan 17 23:50	277.1h	SHORT	$0.4401	$0.3500	+19.34%	−7.64%	+20.13%	take-profit
6	NEAR	Jan 06 19:10	Jan 06 20:47	1.6h	LONG	$1.76	$1.80	+1.66%	−0.71%	+1.81%	take-profit
7	NEAR	Jan 07 13:28	Jan 08 06:15	16.8h	LONG	$1.76	$1.69	−0.38%	−4.09%	−4.26%	stop-loss
8	NEAR	Jan 09 15:23	Jan 12 13:45	70.4h	SHORT	$1.69	$1.66	+1.19%	−4.93%	+1.28%	take-profit
9	PUMP	Jan 10 14:13	Jan 14 16:12	98.0h	LONG	$0.002381	$0.002930	+21.10%	−5.42%	+22.58%	take-profit
10	TRX	Jan 12 13:09	Jan 15 17:07	76.0h	SHORT	$0.2979	$0.3099	−0.29%	−4.43%	−4.45%	stop-loss
11	BTC	Jan 13 11:31	Jan 13 15:59	4.5h	LONG	$92,202.22	$93,400.00	+0.81%	−0.71%	+0.90%	take-profit
12	HYPE	Jan 13 17:21	Jan 13 22:14	4.9h	LONG	$24.46	$25.19	+2.44%	−0.40%	+2.59%	take-profit
13	SOL	Jan 15 13:55	Jan 18 23:46	81.8h	LONG	$145.20	$139.10	−0.14%	−4.58%	−4.59%	stop-loss
14	TRX	Jan 16 17:59	Jan 17 14:15	20.3h	LONG	$0.3080	$0.3144	+1.66%	−0.61%	+1.66%	take-profit
15	SOL	Jan 19 05:26	Jan 20 08:07	26.7h	SHORT	$133.50	$130.40	+1.88%	−1.56%	+1.93%	take-profit
16	POL	Jan 21 17:41	Jan 21 19:59	2.3h	LONG	$0.1329	$0.1376	+3.11%	−0.40%	+3.12%	take-profit
17	TRX	Jan 22 09:16	Jan 28 08:46	143.5h	SHORT	$0.2999	$0.2930	+1.89%	−4.00%	+1.89%	take-profit
18	SOL	Jan 22 09:32	Jan 25 16:01	78.5h	LONG	$129.90	$125.10	+0.20%	−4.05%	−4.08%	stop-loss
19	POL	Jan 22 12:26	Jan 23 14:03	25.6h	LONG	$0.1340	$0.1290	+0.15%	−4.11%	−4.15%	stop-loss
20	FARTCOIN	Jan 26 16:21	Jan 31 14:28	118.1h	SHORT	$0.2824	$0.2300	+17.97%	−11.99%	+18.20%	take-profit
21	SOL	Jan 28 22:15	Jan 29 14:59	16.7h	LONG	$125.46	$120.50	−0.13%	−4.21%	−4.34%	stop-loss
22	TRX	Jan 30 12:19	Jan 31 20:44	32.4h	SHORT	$0.2895	$0.2867	+1.29%	−2.44%	+0.58%	time-exit

Statistics

Metric	Value
Total trades	22
Total PNL	+52.22%
Wins / Losses	15 / 7
Winrate	68%
Mean trade PNL	+2.374%
Std dev per trade	7.676%
Sharpe Ratio (per-trade)	+0.309
Max drawdown (single trade)	−4.59%
Profit factor	2.73
Expectancy per trade	+$2.37 (per $100 stake)

📋 Trade Log with Ollama

Signals

#	Symbol	Opened (UTC)	Closed (UTC)	Held	Dir	Open	Close	Peak	DD	PNL%	Exit
1	SOL	Jan 05 07:13	Jan 05 20:09	12.9h	LONG	$135.23	$139.10	+2.42%	−1.72%	+2.45%	take-profit
2	NEAR	Jan 06 08:46	Jan 06 11:44	3.0h	LONG	$1.77	$1.80	+1.57%	−0.59%	+1.64%	take-profit
3	HYPE	Jan 06 09:59	Jan 06 12:13	2.2h	LONG	$26.52	$27.04	+1.55%	−1.41%	+1.56%	take-profit
4	FARTCOIN	Jan 06 10:43	Jan 17 23:50	277.1h	SHORT	$0.4401	$0.3500	+19.34%	−7.64%	+20.13%	take-profit
5	NEAR	Jan 06 19:10	Jan 06 20:47	1.6h	LONG	$1.76	$1.80	+1.66%	−0.71%	+1.81%	take-profit
6	NEAR	Jan 09 15:23	Jan 12 13:45	70.4h	SHORT	$1.69	$1.66	+1.19%	−4.93%	+1.28%	take-profit
7	PUMP	Jan 10 14:13	Jan 14 16:12	98.0h	LONG	$0.002381	$0.002930	+21.10%	−5.42%	+22.58%	take-profit
8	BTC	Jan 13 11:31	Jan 13 15:59	4.5h	LONG	$92,202.22	$93,400.00	+0.81%	−0.71%	+0.90%	take-profit
9	HYPE	Jan 13 17:21	Jan 13 22:14	4.9h	LONG	$24.46	$25.19	+2.44%	−0.40%	+2.59%	take-profit
10	TRX	Jan 15 12:11	Jan 15 22:35	10.4h	LONG	$0.3055	$0.3124	+1.86%	−0.49%	+1.86%	take-profit
11	SOL	Jan 15 13:55	Jan 18 23:46	81.8h	LONG	$145.20	$139.10	−0.14%	−4.58%	−4.59%	stop-loss
12	TRX	Jan 16 17:59	Jan 17 14:15	20.3h	LONG	$0.3080	$0.3144	+1.66%	−0.61%	+1.66%	take-profit
13	SOL	Jan 19 05:26	Jan 20 08:07	26.7h	SHORT	$133.50	$130.40	+1.88%	−1.56%	+1.93%	take-profit
14	POL	Jan 21 17:41	Jan 21 19:59	2.3h	LONG	$0.1329	$0.1376	+3.11%	−0.40%	+3.12%	take-profit
15	SOL	Jan 22 09:32	Jan 25 16:01	78.5h	LONG	$129.90	$125.10	+0.20%	−4.05%	−4.08%	stop-loss
16	POL	Jan 22 12:26	Jan 23 14:03	25.6h	LONG	$0.1340	$0.1290	+0.15%	−4.11%	−4.15%	stop-loss
17	FARTCOIN	Jan 26 16:21	Jan 31 14:28	118.1h	SHORT	$0.2824	$0.2300	+17.97%	−11.99%	+18.20%	take-profit

Statistics

Metric	Value
Total trades	17
Total PNL	+68.90%
Wins / Losses	14 / 3
Winrate	82%
Mean trade PNL	+4.053%
Std dev per trade	7.918%
Sharpe Ratio (per-trade)	+0.512
Max drawdown (single trade)	−4.59%
Profit factor	6.37
Expectancy per trade	+$4.05 (per $100 stake)

🤖 Baseline VS Ollama

Metric	Without Ollama	With Ollama	Δ
Total trades	22	17	−5 trades skipped
Total PNL	+52.22%	+68.90%	+16.68 pp
Winrate	68%	82%	+14 pp
Wins / Losses	15 / 7	14 / 3	−4 losing trades
Sharpe Ratio	+0.309	+0.512	+0.203
Profit factor	2.73	6.37	+3.64
Expectancy per trade	+$2.37	+$4.05	+$1.68

The LLM correctly vetoed 6 signals, of which 4 were losers (TRX SHORT Jan 06 −4.24%, NEAR LONG Jan 07 −4.26%, TRX SHORT Jan 12 −4.45%, SOL LONG Jan 28 −4.34%) — totaling −17.29% avoided. The two skipped winners (TRX SHORT Jan 22 +1.89%, TRX SHORT Jan 30 +0.58%) cost −2.47%. Net win: +14.82% from filtering, plus one additional trade (TRX LONG Jan 15 +1.86%) the LLM run captured.

🛠️ Architecture in 60 seconds

Layer	File	Responsibility
Crawl	`packages/core/src/lib/services/core/CrawlerService.ts`	Telegram `iterMessages` → `parser-items` upsert
Parse	`packages/core/src/lib/services/screen/ChannelScreenService.ts`	Regex extraction of `direction/entry/targets/stoploss`
Job	`packages/core/src/lib/services/job/SignalJobService.ts`	Subscribes to `signalJobSubject`, runs each unvisited row through `SignalLogicService`
Outline	`packages/core/src/logic/outline/risk.outline.ts`	LLM prompt + `commitMetricsHistory` + zod-validated response
Logic	`packages/core/src/lib/services/logic/SignalLogicService.ts`	Pure passthrough of LLM verdict to `screen-items` DTO
Schema	`packages/core/src/schema/Screen.schema.ts`	Final stored fields: `direction`, `entryFrom/To`, `targets`, `stoploss`, `riskAction`, `riskSureLevel`, `riskConfidence`, `riskDescription`, `riskReasoning`
Strategy	`content/jan_2026.strategy/jan_2026.strategy.ts`	Reads `screen-items`, opens position only when `riskAction === "follow"`
Cron	same file, `Cron.register(...)`	`interval: "15m"` for live polling; no `interval` = one-shot backtest prepare

🔑 Environment Variables

Var	Default	Purpose
`CC_MONGO_CONNECTION_STRING`	`mongodb://localhost:27017/backtest-pro?wtimeoutMS=15000`	Mongo connection
`CC_TELEGRAM_API_ID`	`31861455`	Telegram MTProto API ID
`CC_TELEGRAM_API_HASH`	`ca60446c67ce250ee4e789c730163449`	Telegram MTProto API hash
`CC_SYMBOL_LIST`	`BTCUSDT,POLUSDT,ZECUSDT,HYPEUSDT,DOGEUSDT,SOLUSDT,PENGUUSDT,TRXUSDT,HBARUSDT,NEARUSDT,FARTCOINUSDT,ETHUSDT,PUMPUSDT,ENAUSDT`	Comma-separated symbols crawled per run
`CC_OLLAMA_URL`	`http://localhost:11434`	Ollama HTTP endpoint serving `gpt-oss`

📜 Summary

Aspect	This monorepo
LLM gate	Local Ollama (`gpt-oss` quantized) consulted per signal, returns `riskAction: "skip" \| "follow"`
Empirical rules	Embedded in LLM system prompt (`avgRangePct < 0.07%` for SHORT; `momentum24hPct < -1%` for LONG) — tunable without recompiling `packages/`
Live ingestion	15-minute crontab via `Cron.register(.., interval: "15m")` polls the configured channel
Backtest result	January 2026: +52.22% → +68.90% with the LLM gate (Sharpe +0.309 → +0.512, winrate 68% → 82%)
DI surface	`globalThis.core` typed via root `tsconfig.json` paths → rolled-up `types.d.ts`
Strategy isolation	Strategy files in `./content/` are loaded by `@backtest-kit/cli` at runtime, never bundled into `@pro/*`

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
assets		assets
config		config
content/jan_2026.strategy		content/jan_2026.strategy
docker		docker
modules		modules
packages		packages
scripts		scripts
.env.example		.env.example
.gitignore		.gitignore
README.md		README.md
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

🧿 backtest-ollama-crontab

🚀 Quick Start

1. Build the workspace packages

2. Authenticate the Telegram MTProto session

3. Copy the session into the strategy folder

4. Run the backtest

🎯 What this repo demonstrates

Pipeline

📊 Before vs After Ollama

📋 Trade Log WITHOUT Ollama

📋 Trade Log with Ollama

🤖 Baseline VS Ollama

🛠️ Architecture in 60 seconds

🔑 Environment Variables

📜 Summary

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

🧿 backtest-ollama-crontab

🚀 Quick Start

1. Build the workspace packages

2. Authenticate the Telegram MTProto session

3. Copy the session into the strategy folder

4. Run the backtest

🎯 What this repo demonstrates

Pipeline

📊 Before vs After Ollama

📋 Trade Log WITHOUT Ollama

📋 Trade Log with Ollama

🤖 Baseline VS Ollama

🛠️ Architecture in 60 seconds

🔑 Environment Variables

📜 Summary

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages