Skip to content

Setup

Prerequisites

  • pyenv (Run curl https://pyenv.run | bash)
  • nvm (Run curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash)
  • ripgrep
  • gcloud

Installation

git clone https://github.com/TrafficGuard/nous.git
cd nous
source ./bin/configure
The configure script will:

  • Ensure the python version in .python-version is installed and install aider.
  • Ensure the node.js version in .nvmrc is installed and run npm install
  • Initialise the environment variable file at variables/local.env
  • Change to the frontend folder and run npm install

When enabled Google Cloud is used for Firestore database persistence, the Gemini AI models, Anthropic Claude on Vertex, and tracing via OpenTelemetry.

A Dockerfile is also provided to deploy application in Cloud Run or on a VM with the container-optimised OS.

  • Create a project in Google Cloud and install the gcloud command line tool.
  • In variables/local.env update the GCLOUD_PROJECT and GCLOUD_REGION variables.
  • Run ./bin/gcp_setup which will:
  • Run gcloud auth application-default login which provide credentials for the Google Cloud SDKs. (If the webpage fails to load then ensure port the callback webpage opens with isn't in use)

To use Anthropic Claude through the Vertex API you will need to enable the Claude models from the Model Garden. Make sure to click Accept on the final screen.

Model garden links - 3.5 Sonnet - 3.0 Haiku

As Claude is only available in select regions there is an additional environment variable GCLOUD_CLAUDE_REGION in the sample local.env file which defaults to us-east5

Non-Google Cloud setup

If you want to get running ASAP then in variables/local.env update the DATABASE variable to memory.

Configuration quick start

The variables/local.env file contains the configuration when running Nous locally.

If you have configured Google Cloud then update TRACE_AGENT_ENABLED to true.

By default, Nous runs in single_user authentication mode. A user profile will be created the first time the application is run. Update the SINGLE_USER_EMAIL variable with your email before running Nous for the first time.

The LLM service API keys and integration configurations can be set in web UI, or alternatively in the variables/local.env file. Values in the user profile take preferences over the environment configuration values.

Quick links to create API keys:

LLMs - Anthropic - OpenAI - Groq - Together.ai - Fireworks.ai - Deepseek

Integrations - Perplexity - Jira - GitLab - GitHub

For observability configuration (logging and tracing) see the Observability page.

Running quick start

CLI

In the src/cli folder are the file gen.ts, agent.ts, swe.ts, code.ts, research.ts, util.ts which are scripts that can be run with npm run <script> <optional_prompt>.

If a prompt isn't provided, then it will read the prompt from the file indicated. This gives the option of quick inputs and more prepared, structured inputs.

  • npm run gen does a single text generation from an LLM, if no arguments reading the input from src/cli/gen-in
  • npm run agent runs the autonomous agent, using the functions configured in src/cli/agent.ts, and if no arguments reading the input from src/cli/agent-in
  • npm run code runs the Code Editing agent on the repository, if no arguments reading the input from src/cli/code-in
  • npm run swe runs the Software engineer agent, which can find a remote repo to clone, edit and create a pull/merge request. If no arguments reading the input from src/cli/swe-in

Full details at Running from the CLI

Local server & UI

In one terminal run 

npm run start:local
In a second terminal run 
cd frontend
npm run start:local
The UI will be available at http://localhost:4200

Full details at Running from the UI

Documentation for deploying on Google Cloud will be provided soon.

Tests

Running tests

Keep the Firestore emulator running in a separate shell or in the background 

gcloud emulators firestore start --host-port=127.0.0.1:8243

npm run test