Socratic: KnowledgeOps for Expert-Level Domain Agents

Socratic is a KnowledgeOps platform that helps teams build expert-level knowledge base for vertical, domain-specific agents. You point Socratic at your existing domain docs, and Socratic collaborates with you to build a curated and reliable knowledge base that you actually trust.

Socratic is a KnowledgeOps platform that helps teams build expert-level, trustworthy knowledge bases for vertical, domain-specific agents.
Point Socratic at your existing docs, and collaborate with Socratic to synthesize a curated, reliable, auditable knowledge base.

Overview

High-quality domain knowledge is critical for building high-value agents.

Today, most teams rely on "document dump" workflows (e.g., RAG): give the agent all the files and hope it interprets them correctly.
But this fails for the same reason it would with a human - you wouldn't onboard a new hire by handing them a folder of docs and saying "good luck."

Socratic rethinks how expert knowledge should be distilled. It treats knowledge building like teaching a human expert: capturing tacit rules, resolving ambiguities, surfacing edge cases, and clarifying domain intuitions.

Socratic does this by combining:

• Interactive collaboration with the human expert
  Socratic asks high-quality questions, identifies ambiguities, and requests clarification.
• Grounded research over your actual domain documents
  It reads your code, specs, and docs to extract accurate conceptual models.
• Knowledge as a first-class artifact
  It generates clean, modular, plain-text knowledge units - fully auditable and directly usable by downstream agents.
• Human-in-the-loop curation
  Every proposed change can be inspected, diffed, refined, and explicitly approved.

The result is a structured, trustworthy knowledge base, not an opaque embedding index giving you control over what your agent actually knows.

Demo

Using Socratic to build knowledge base to understand Socratic itself (3-min): https://youtu.be/R4YpbqQZlpU

Using Socratic to build knowledge base for Google Analytics SQL agent: https://youtu.be/L20vOB3whMs

Install

Linux kernel >= 5.15 is required.

git clone https://github.com/kevins981/Socratic.git
cd socratic

# optional
conda create -n socratic python=3.10 -y
conda activate socratic

# install
pip install -e .

Install OpenAI Codex:

# might need sudo
npm install -g @openai/codex

For Web-UI:

# if npm is not installed
sudo apt install nodejs npm
cd web
npm install

Running

Assume that the project name is airline_demo and relevant source files are located in examples/repos/tau_airline.

Web UI (recommended):

# 0. Create project
socratic-cli create --name airline_demo --input_dir examples/repos/tau_airline 

# 1. Launch web UI
cd web
npm run dev:project -- --project airline_demo

Command line interface:

# Create project
socratic-cli create --name airline_demo --input_dir examples/repos/tau_airline

# Start working on your knowledge base!
socratic-cli synth --project airline_demo

# Digest model to look deeper into your existing knowledge base
socratic-cli digest --project airline_demo

# Export knowledge base (Agent.md)
socratic-cli export --project airline_demo

Supported Models

To configure the LLM model to use, create a .env file at the project root (see .env.example).

OpenAI Models

Socratic is powered by OpenAI Codex. If you already have access to Codex via a ChatGPT subscription, you can directly use it by setting PROVIDER=chatgpt in .env file. See .env.example for examples.

• Note that the set of models supported by ChatGPT Codex is limited. See https://developers.openai.com/codex/models/.

You can also bring your own OpenAI API key and use any OpenAI model available through the API. To do this, specify PROVIDER=openai, BASE_URL=https://api.openai.com/v1, and ENV_KEY=OPENAI_API_KEY. Make sure OPENAI_API_KEY env var is set in your shell.

OpenRouter Models and Local Models

In addition, Socratic works with any LLM that exposes an OpenAI-compatible API, including locally hosted models. See .env.example for examples.

Notes on Model Quality

Some models do not work well with Socratic due to tool-calling issues. In practice:

• Most stable: gpt-5.2(-codex), gpt-5.1
• Works well: glm-4.6, grok-4

In some cases, the model may encouter tool calling problems and therefore fail unexpectedly. E.g., gpt-oss-120 is known to be unstable. In general, smaller models tend to be more unstable. Socratic relies on multi-step reasoning and tool operations, so the model must be sufficiently capable to run the full pipeline reliably.

Privacy & Security

• Local storage: All files and outputs are stored entirely on your own machine. Socratic does not upload, transfer, index, or store your data anywhere else.
• Local processing: All analysis and processing happen locally, except when data is sent to an external LLM provider (e.g., OpenAI) using your own API key.
• Sandboxed terminal agent: Socratic uses Codex as its terminal agent to read and analyze source documents. Socratic runs Codex in read-only mode, preventing the agent from editing files or running commands that require network access. See the Codex sandbox documentations for more details.