ChainPeer

ChainPeer v1.0 is now available

The intelligent terminal
for modern engineering.

A universal, lightweight, and incredibly capable coding agent. Built to understand context, write code, and solve problems directly in your terminal.

Try Install

Meet ChainPeer

Your AI engineering partner.

Plan & Execute

ChainPeer treats planning as executable control state, not loose notes. The plan tools turn complex work into inspectable steps, dependency edges, and strict progress updates.

DAG task design: steps declare depends_on, parallel work stays parallel, and blocked dependencies cannot be executed early.

Plan tools coordinate execution: plan_create, plan_next, plan_update_step, plan_link_dependency, and plan_close provide scheduling, cycle checks, state transitions, and expected_version optimistic locking.

>_ ChainPeer v0.1.1

Ready for engineering work

model glm-5-turbo|workspace test

Type /help|Enter send|Ctrl+J newline

Ctrl+O history|Ctrl+L clear|Ctrl+C drafts

首先，创建计划：

Tokens: input 5.8k / 245.5k (2.4%), cached 0 (0.0%), output 452

Running plan_create

Tool: plan_create completed in 2ms

计划已创建，版本号为 1。现在查看可执行的步骤并开始工作。

Tokens: input 6.1k / 245.5k (2.9%), cached 5.7k (80.0%), output 52

Running plan_next

Tool: plan_next completed in 1ms

Step 1 可以开始。先创建 data 目录，然后创建三个 Python 文件（它们之间的代码无实际依赖，可以一起创建）。

Tokens: input 7.4k / 245.5k (3.0%), cached 7.1k (96.6%), output 452

Running bash: mkdir -p data

Running plan_update_step

Tool: bash completed in 146ms

Tool: plan_update_step completed in 10ms

现在创建三个 Python 文件：

Tokens: input 7.7k / 245.5k (3.1%), cached 5.7k (73.8%), output 1.3k

Running write_file

Tool: write_file completed in 1ms

三个文件已创建完成。现在将前三个步骤标记为完成，并开始执行。

Tokens: input 9.3k / 245.5k (3.8%), cached 7.7k (82.7%), output 117

Running plan_update_step

Tool: plan_update_step completed in 10ms

Long Context & History Recovery

Never lose track. Seamlessly resume long conversations with intelligent history summarization.

>_ ChainPeer v0.1.1

Ready for engineering work

model glm-5-turbo|workspace test

Type /help|Enter send|Ctrl+J newline

Ctrl+O history|Ctrl+L clear|Ctrl+C drafts

Resumed session 20260604_1...fb38: 16 visible message(s), showing last 6.

Full context is still loaded; 10 older message(s) are hidden from the terminal.

- assistant: 任务1验证通过，10个随机数已写入 JSON。现在执行任务2。

- assistant: 任务2执行成功。验证输出。

- assistant: 统计数据验证通过。现在执行最后一个任务3。

- assistant: 任务3执行成功。读取最终报告并呈现。

- assistant: 关闭最后一个步骤和计划。

- assistant: 全部完成。下面是整个流程的总结和最终报告。 --- ## 任务链式依赖执行总结 ### 三个任务及其依赖关系 ```任务1 > 任务2 > 任务3```

↓ random_numbers.json → statistics.json → final_report.txt ``` ### 任务说明 | 任务 | Python 文件 | 输入 | 输出 ...

Use /status for details, /sessions to switch, or continue typing below.

You > |

session 20260604...fb38 | model glm-5-turbo | ctx 6.1% cache 38.5%

Engineered for precision.

Every detail of ChainPeer is crafted to eliminate friction between your thought process and the final code.

Universal Context

ChainPeer seamlessly indexes your entire codebase, understanding the subtle architectural decisions and patterns unique to your project. It writes code that looks like yours.

Lightning Fast

Native execution with zero bloat. ChainPeer runs locally and connects to models with minimal latency.

Multi-Model

ChainPeer currently targets OpenAI-compatible API models. Its model-side interaction design and prompts are primarily tuned around the GLM model family.

Terminal Native

No context switching. Run tests, review diffs, and execute complex refactors without ever leaving your command line interface.

Industry Innovation

Workflow Design

The core architectural breakthrough of ChainPeer.

The model understands intent, explains complex inputs, and chooses the right workflow.

Workflow turns repeated work into deterministic commands, schemas, preflight checks, runs, and artifact contracts.

ChainPeer provides planning, context, session recovery, and tool execution; ChainPeerFi injects finance prompts and workflow commands at runtime.

Financial Agent Extension

ChainPeerFi

A specialized financial agent extension for quantitative research and investment analysis.

Factor Detection & Research

Check data, define factor logic, build reproducible experiments, and report leakage, bias, and robustness risks.

>_ ChainPeerFi v0.1.1

Finance research and strategy workspace

model glm-5-turbo|workspace test

Type /help|Enter send|Ctrl+J newline

Ctrl+O history|Ctrl+L clear|Ctrl+C drafts

You > 请使用 $factor_detection 对 (-1 * Ts_Rank(rank(low), 9)) 进行因子回测

Agent:

Skill: factor_detection (explicit_dollar_name)

我将按照因子检测工作流来执行回测。首先，让我解读因子定义并检查本地可用数据。

因子定义解读：

- rank(low)：横截面排名（按当日最低价在全市场中的百分位排名）

- Ts_Rank(..., 9)：在 9 天滚动窗口内的时序排名

- -1 *：取反，即做多时序排名低的标的（近期最低价排名处于低位的标的的）

让我先检查本地数据和项目结构。

Tokens: input 7.2k / 245.5k (2.9%), cached 0 (0.0%), output 349

Running bash: python -m chainpeer_fi.commands.finance factor discover-data

Running bash: python -m chainpeer_fi.commands.finance factor preflight --config factor.json

Tool: bash completed in 146ms

预检通过。现在执行因子回测并生成 manifest、指标文件和 HTML 报告。

Running bash: python -m chainpeer_fi.commands.finance factor run --config factor.json

20260604_170344_neg_ts_rank_rank_low_9

Each run writes an auditable artifact folder: charts, config, structured metrics, Markdown notes, PDF report, and manifest for reproduction.

config.json fixes the factor expression, data scope, and backtest parameters.

factor_report.json stores structured metrics for programmatic comparison.

factor_report.md and factor_report.pdf provide readable research notes and a fixed-format report.

manifest.json records run artifacts, commands, and reproduction metadata.

charts/ outputs IC, quantile returns, drawdown, turnover, and sample-stability visuals.

deterministic

charts/13 png

cumulative_rank_ic.png

factor_distribution.png

factor_vs_forward_return.png

ic_decay.png

long_short_drawdown.png

monthly_rank_ic_heatmap.png

quantile_cumulative.png

quantile_mean_returns.png

rank_ic_timeseries.png

rank_ic_vs_sample_size.png

top_bottom_turnover.png

walk_forward_rank_ic.png

yearly_rank_ic.png

config.json

JSON

factor_report.json

JSON

factor_report.md

factor_report.pdf

PDF

manifest.json

RUN

Fixed-format PDF factor report

ChainPeerFi can consistently output the same PDF report structure for every factor backtest, making results easy to compare, archive, and review.

Strategy Iteration

Manage experiment configurations, validate runs, and enable reproducible strategy improvements.

>_ ChainPeerFi v0.1.0

Finance research and strategy workspace

model glm-5-turbo|workspace test

Type /help|Enter send|Ctrl+J newline

Ctrl+O history|Ctrl+L clear|Ctrl+C drafts

You > 请使用 $long_strategy_iteration对该项目下的策略进行迭代，目标：去手续费年化>30%，最大回撤<15%。

Agent:

Skill: long_strategy_iteration (explicit_dollar_name)

我先来探索一下项目结构并理解现有的策略和数据。

Tokens: input 6.6k / 245.5k (2.7%), cached 0 (0.0%), output 93

Running list_files: .

Tool: list_files completed in 2ms

Tokens: input 7.0k / 245.5k (2.9%), cached 6.6k (93.8%), output 62

Running read_file: data/final_report.txt

Running read_file: data/statistics.json

Running read_file: data/reproduce_config.json

Running read_file: data/sample_reproduce_config.json

Tool: read_file completed in 7ms

Tool: read_file completed in 0ms

Tokens: input 10.9k / 245.5k (4.4%), cached 7.0k (64.7%), output 94

Running read_file: outputs/research_extraction.md

Running read_file: outputs/data_requirements.md

Broker Report Reproduction

Convert formulas, tables, and conclusions into auditable code.

>_ ChainPeerFi v0.1.0

Finance research and strategy workspace

model glm-5-turbo|workspace test

Type /help|Enter send|Ctrl+J newline

Ctrl+O history|Ctrl+L clear|Ctrl+C drafts

You > 请使用 $research_reproduction对 'xxxx研报.pdf' 进行复现。

Agent:

Skill: research_reproduction (explicit_dollar_name)

我来按照 research_reproduction 的工作流，首先在项目目录中定位并读取该研报 PDF。

Tokens: input 6.6k / 245.5k (2.7%), cached 0 (0.0%), output 113

Running glob: **/*研报*.pdf in .

Running glob: **/*.pdf in .

Tool: glob completed in 1ms

找到文件 xxxx研报.pdf (7.4 MB)。先读取前 30 页内容以提取研报核心信息。

Tokens: input 6.9k / 245.5k (2.8%), cached 6.6k (95.6%), output 58

Running read_pdf: xxxx研报.pdf

Tool: read_pdf completed in 39.13s

研报内容已提取完毕。现在让我整理关键信息并检查工作环境。

Tokens: input 12.2k / 245.5k (5.0%), cached 6.9k (56.8%), output 623

Running bash: cd /e/code/agent/test && python -m chainpeer_fi.commands.finance --help 2>&1 | head -50

Running bash: cd /e/code/agent/test && python -m chainpeer_fi.commands.finance factor

Investment Analysis

Separate assumptions based on provided facts, explain risks, and avoid hallucinated data.

>_ ChainPeerFi v0.1.0

Finance research and strategy workspace

model glm-5-turbo|workspace test

Type /help|Enter send|Ctrl+J newline

Ctrl+O history|Ctrl+L clear|Ctrl+C drafts

You > 请使用 $investment_advice，结合我的个人持仓情况（'持仓情况.txt'），进行投资分析

Agent:

Skill: investment_advice (explicit_dollar_name)

我先读取您的持仓文件，了解您的投资情况。

Tokens: input 6.6k / 245.5k (2.7%), cached 6.3k (94.7%), output 54

Running read_file: 持仓情况.txt

Tool: read_file completed in 0ms

现在让我检查一下投资分析工具是否可用。

Tokens: input 7.5k / 245.5k (3.0%), cached 6.6k (88.4%), output 1.2k

Running bash: python -m chainpeer_fi.commands.finance investment --help 2>&1 | head -30

接下来将结合持仓结构、风险暴露和可用市场数据生成投资分析。

Try ChainPeer

For local engineering work, planned execution, and long conversation recovery.

Try Install

Consult ChainPeerFi

For quantitative research, factor backtesting, strategy iteration, and investment analysis.

contact@sky-castle.com.cn

ChainPeer

The intelligent terminal for modern engineering.

Meet ChainPeer

Plan & Execute

Long Context & History Recovery

Engineered for precision.

Universal Context

Lightning Fast

Multi-Model

Terminal Native

Workflow Design

The core architectural breakthrough of ChainPeer.

ChainPeerFi

Factor Detection & Research

Fixed-format PDF factor report

Strategy Iteration

Broker Report Reproduction

Investment Analysis

Try ChainPeer

Consult ChainPeerFi

The intelligent terminal
for modern engineering.