Model or dataset
tgoai/tgo avatar
tgoai/tgo

tgo: fifteen sub-repositories, one bootstrap script, and twelve services by default

Open-source AI Agent Customer Service Platform. Build AI agent teams with LLM orchestration, RAG knowledge base, multi-channel support, and human collaboration.

664 stars126 forksTypeScriptNOASSERTION

At a glance

What is it?
tgo is an open-source customer service agent platform that bundles fifteen sub-repositories behind one Compose stack: an admin frontend, seven Python services, a retrieval service on pgvector, an instant messaging server and five widget SDKs. The details worth reading are in the Makefile and the environment files rather than in the feature list.
Who is it for?
tgo suits a team that wants to self-host customer service with retrieval and human handoff rather than embed a commercial widget, and that is prepared to run fifteen repositories and a dozen services. Before you deploy, change the database credentials, because the example environment ships the same string for the database name, the user and the password while declaring a production environment.
Can I use it commercially?
Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
Is it still maintained?
Activity is slowing. The repository last received commits 6 months ago.
What is it written in?
Mainly TypeScript, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • tgoai/tgo664 ★ · +54
Star history
Repository2026-09-112026-11-15
tgoai/tgo610664

54 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

One command pipes a remote script into your shell

The recommended install is a single line, and it executes code fetched at the moment you run it:

bash
REF=latest curl -fsSL https://raw.githubusercontent.com/tgoai/tgo/main/bootstrap.sh | bash

The script is described as checking requirements, cloning the repository and starting the services, and the shown `REF=latest` assignment applies to `curl`, not the `bash` process after the pipe. To pass a chosen value to the script, export it in the shell or assign it to `bash` explicitly. A second script exists for users inside China, hosted on a Gitee mirror rather than GitHub, and the repository carries a separate Compose file for that path as well as a generator script that appears to build the bootstrap files. Two details are worth noting before you run any of it. The reference is pinned to the main branch of a moving repository rather than to a tag, and a moving branch is not an immutable reference; review and pin a commit when reproducibility is required. Local development avoids the remote script entirely: copy a development environment file and run a make target.

Fifteen repositories across six languages

The stack is a monorepo of services, and the table gives each one a job and a language. Seven are Python services on an ASGI framework: the operations service that manages agents, tool bindings, knowledge bases and usage analytics; the core service handling users, visitors, assignment and messaging; the retrieval service; the plugin runtime; the device control service; the channel intake service for WeChat, Feishu, DingTalk, Telegram, Slack and email; and the workflow engine, which executes a directed acyclic graph of model, API, condition and tool nodes. Two more are TypeScript: the admin frontend and a command line tool that doubles as a model-context-protocol server with more than forty built-in tools. One is Go. The widget side adds five SDKs in TypeScript, Swift and Dart, and the web widget pins React 18 while the admin frontend is on React 19.

The Makefile starts twelve application services by default

Local development runs through a Makefile whose default target prints help rather than starting anything. The dev target merges two Compose files, the base one and a development overlay, against an environment file whose name defaults to the development copy. Three services count as infrastructure: Postgres with the vector extension, Redis with persistence enabled, and the messaging server. Then there are seven services that need building and twelve that run, which is the number to plan around, since the workers and the scheduled beat jobs are part of the default set rather than something you opt into. Two profiles add more: a monitoring profile with a queue dashboard and a database browser, and an agent profile that adds one device service. A disable variable removes named services, and the readme shows it used to skip two workers.

The example database password is the database name

The example environment file sets three values to the same string: the database name, the user and the password are all `tgo`, and the port is the Postgres default. Nothing in that file suggests the values must be changed, while the same file declares the environment as production. The Compose file takes the same variables with those defaults repeated, so a deployment that copies the example unchanged runs with a password anyone can guess from the project name. The rest of the storage configuration is more considered. Data directories for Postgres, Redis and the messaging server are bind-mounted into a directory inside the project rather than left as anonymous volumes, so deleting the stack does not delete your data and a backup is a directory copy. Storage backends beyond local disk are also offered through a type setting.

Only the reverse proxy is meant to be published

The environment file maps out two port groups. The frontend and backend ports are labelled as internal, reachable only from other services inside the Compose network, and the proxy is the single published entry point on the standard web ports. The example file leaves three domain fields empty, which per its own comments means you are using direct port access rather than name-based routing, and the certificate mode defaults to none with an automatic option that requests a certificate from a public authority and keeps renewing it. A body size limit of a hundred megabytes is set on the proxy, which is the number to revisit if you intend to send documents through the widget rather than text. Store endpoints point at a hosted domain under the project's own name, so a self-hosted install is making at least one outbound call unless you change them.

A dated messaging image and three published ports

The instant messaging service is pinned to a build rather than to a plain release: its image reference carries a version and a date stamp, so an upgrade means the project's maintainers rebuilding rather than you choosing. Its cluster settings are hard-coded inside the Compose file as internal and fixed, including a node identifier and internal addresses for the cluster and TCP ports, while the external addresses come from the environment with a loopback default. Three ports are published for it, one for TCP traffic, one for websockets and one for its own administration interface, which is the one to firewall if you deploy this anywhere reachable. Token authentication for the messaging server defaults to off. The service also posts events to a webhook path on the core API service, which is how message traffic reaches the agent pipeline.

One service exists to run shell commands on managed hardware

Two of the fifteen repositories deserve a second look. The device agent is written in Go and runs on managed devices, providing file and shell capabilities over a JSON-RPC transport. Its partner, written in Python, manages those connections on the server side and is described as carrying a model-context-protocol agent. Between them they give a remote agent the ability to touch a filesystem and run commands on hardware, which is a different category of capability from everything else in this platform, where the other services handle chat, retrieval and routing. The rest of the root is infrastructure the readme does not mention: two shell scripts, a skills directory, four Compose files including a tools-only one, a stream usage guide, and separate directories for the documentation site and the marketing site. The licence situation is similarly split, with a main licence file and a Chinese-language counterpart.

Editorial conclusion

tgo suits a team that wants to self-host customer service with retrieval and human handoff rather than embed a commercial widget, and that is prepared to run fifteen repositories and a dozen services. Before you deploy, change the database credentials, because the example environment ships the same string for the database name, the user and the password while declaring a production environment. Decide on your exposure path too, since only the reverse proxy is meant to be published and the rest is reachable only inside the Compose network. And read the device services before enabling them, because one of them exists to run shell and file operations on managed hardware.

Frequently asked questions

What does the tgo platform include?

Fifteen sub-repositories: seven Python services including retrieval, workflow and channel intake, a TypeScript admin frontend, a Go device agent, a TypeScript command line tool that doubles as an MCP server, and five widget SDKs for web, iOS, Flutter, a mini program and the command line.

How do I run tgo locally for development?

Copy the development environment file and run the dev target, which merges the base and development Compose files. Profiles add monitoring or agent services, and a disable variable removes named services from the default set of twelve.

Which ports does tgo expose to the host?

The reverse proxy on the standard web ports. The service ports for the API, the AI service, the platform, retrieval and the web frontend are internal to the Compose network, and the messaging server publishes separate ports for TCP, websockets and its admin interface.

What hardware does tgo need?

At least four CPU cores and 8 GiB of RAM, on macOS, Linux or WSL2.

Official sources

  1. Issues
  2. Project website
  3. README
  4. tgoai/tgo on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/tgoai-tgo.svg)](https://hysenlabs.com/projects/tgoai-tgo)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/tgoai-tgo.svg)](https://hysenlabs.com/projects/tgoai-tgo)

Was this useful?