StoryCodeThe AI coding agent that remembers your project's story

The "why" behind every change is captured as a Story and recalled automatically in new sessions, so your agent never starts from zero. Runs locally, with any model you choose.

Download StoryCodeExplore capabilities

Available for macOS · Windows · Linux

PROJECT MEMORY

Why the name StoryCode?

No manual note-taking. Key decisions made while the agent works become Stories naturally, so the next session starts with your project's history instead of a blank slate.

Automatic checkpoints

Code changes are linked to Git commits automatically — recording not just what changed, but why.

Session archiving

Great sessions become reusable knowledge snippets in one click. Every save is versioned and full-text searchable.

Linked and traceable

Stories link to sessions, files, Git commits, and each other — requirements ↔ code ↔ decisions, all traceable.

Auto Recall

New sessions automatically load relevant Stories as context, so the agent starts with your project's memory.

Without Story

Every new session means re-explaining the context, the pitfalls, and the trade-offs you already made.

With Story

The agent recalls the relevant decisions and picks up right where you left off.

CAPABILITIES

An agent workbench built for real development workflows

01HANDS-ON

Hands-on: code, terminal, and Git

The agent doesn't just suggest — it edits files, runs commands, and manages Git changes, always within the boundaries you set.

  • Reads and writes project files, understands code via tree-sitter
  • Built-in terminal, reviewable Git changes
  • OS-level sandbox + human approval for risky operations
02KNOWLEDGE

Local RAG knowledge base, ready out of the box

Turn project docs, domain material, and session conclusions into searchable knowledge the agent cites when it answers.

  • Bundled embedding and OCR models — no downloads needed
  • Incremental indexing for Markdown, PDF, Word, Excel, PowerPoint, images, and more
  • Hybrid full-text + vector search; archive great answers back in one click
03AUTOMATION

The agent keeps working while you rest

Package repetitive work as workflows, schedule them, and direct the agent from any chat app.

  • Parameterized workflows with cron scheduling, pause/resume, and run history
  • Parallel subagents, background runs, and resumable long tasks
  • 9 IM platforms including WeChat, Feishu, DingTalk, Telegram, Discord, and WhatsApp

And much more

Local inference

Built-in llama.cpp with hardware-aware model recommendations. Fully offline.

MCP tool ecosystem

A pluggable tool system built on MCP. Install extensions and go.

Computer Use

UI automation, web scraping, and PDF / Word / Excel processing.

Web search

Tavily / Brave / Bocha / DuckDuckGo, with cited sources.

Interactive visuals

Mermaid, Sankey, radar charts, and maps from a single prompt.

Local voice

Whisper / Qwen3 dictation and Kokoro / Qwen3 text-to-speech.

Media generation & capture

Text-to-image, image-to-video, screen recording, and screenshots.

Databases & remote hosts

Query SQLite / PostgreSQL / MySQL, manage SSH / SFTP, keep credentials in a vault.

SCENARIOS

What StoryCode Can Do for You

Write code, research, handle documents, get scheduled reports — a few concrete, relatable workflows

CODE

Ship a feature end to end

Describe the task. The agent understands the codebase, edits files, and runs tests in the terminal. Changes land as a Git diff for you to review before committing, and the "why" is saved to Story.

STORY

Project memory: captured automatically, or written by you

Key information from the agent's work is captured as Stories automatically, and you can write your own — requirement context, lessons learned, team conventions. New sessions automatically recall them and inject them into the context; when an old bug resurfaces, the earlier diagnosis can be pulled up right away; and throughout the project's lifecycle, you can trace the reasoning behind every decision at any time.

DEVTOOLS

Databases and remote hosts in one window

Query and edit SQLite / PostgreSQL / MySQL and export to CSV; manage remote hosts over SSH and transfer files via SFTP, with credentials kept safe in the Vault.

KNOWLEDGE

Turn documents into a searchable knowledge base

Drop in PDFs, Word, Excel, PPT, or images to index them automatically. The agent searches them when answering and cites its sources.

RESEARCH

Research the web, get charts and conclusions

Search the web with cited sources, then turn the results into a Mermaid diagram, radar chart, or Sankey chart in one sentence.

WORK

Documents and spreadsheets

Batch-clean Excel sheets, rewrite Word docs, extract PDF content, or let the agent operate UIs and scrape web pages.

MEDIA

Images, video, and voice

Screen recording, screenshots, and audio recording; image generation and video generation.

SCHEDULE

Scheduled tasks that run themselves

Package repetitive work into a workflow on a schedule — e.g. summarize yesterday's changes every morning and push them to WeChat / QQ / Feishu.

IM

Command the agent from chat apps

On the go? Control StoryCode remotely from WeChat / QQ / Feishu to have the agent on your computer do the work and report back.

MODEL FREEDOM

Model freedom: cloud, subscriptions, and local

Your keys, your quota. Switch between 20+ providers and gateways, or run fully offline with built-in local inference.

MODEL PROVIDERS
14
OpenAIAnthropicGoogle GeminixAI GrokGroqDeepSeek智谱 GLM阿里百炼 QwenKimiMiniMax火山引擎方舟小米 MiMo美团 LongCatMistral
CLOUDS & GATEWAYS
7
OpenRouterAzureAWS BedrockGCP Vertex AILiteLLMSnowflakeDatabricks
AGENT / CLI SUBSCRIPTIONS
5
CodexClaude CodeCursor AgentGemini CLIGitHub Copilot
LOCAL INFERENCE
1
llama.cppbuilt-in
PRIVACY & CONTROL

Your code and documents stay on your machine

Data stays local

Code, prompts, knowledge, and sessions are stored locally by default; only the model provider you choose receives conversation content. The web account only handles downloads, subscriptions, and device binding.

Sandboxed execution

Terminal commands run inside OS-level sandboxes (macOS Seatbelt / Linux Landlock / a dedicated Windows sandbox). Choose read-only, workspace-write, or full access, and control network access separately.

Asks before risky actions

Dangerous commands and prompt injection are checked before execution. High-risk operations wait for your confirmation, and approval policies can be adjusted as needed.

No collection, no leaks

StoryCode never collects your code, prompts, files, or private configuration; API keys stay on your local machine, and other credentials are kept in the Vault.

Give your agent a memory of your project

Download StoryCode, bring your own API key or a local model, and start your first session in minutes.

Download StoryCodeRead the user guide

macOS · Windows · Linux

StoryCode - Local-first AI Coding Agent Workbench