PacketChatDocs

PacketChat#

PacketChat is a self-hosted, multi-user workspace for chat, reusable prompts, projects, knowledge bases, reusable agents, and centrally managed model-provider accounts. An organisation runs the service and controls the provider keys; users sign in to the shared installation.

What it includes#

The deployed stack provides:

  • streamed model conversations with persisted branches, file attachments, downloadable exports and revocable sharing links;
  • personal projects and versioned prompt templates;
  • document ingestion and knowledge search;
  • reusable agents with published versions and approval-gated actions;
  • private-project PacketAgent connections and threaded Worker run cards;
  • administrator-managed users, provider accounts, model discovery, usage, and audit views; and
  • PostgreSQL, Redis, and S3-compatible object storage dependencies in the provided Compose stack.

PacketChat supports OpenAI, Azure OpenAI, Anthropic, Google Gemini, xAI, Sugar, OpenCode Go, Ollama Cloud, ClinePass, Perplexity, MiniMax and custom OpenAI-compatible provider accounts. Availability of an individual model depends on the provider account and its discovery response.

Operating model#

Provider credentials belong to the installation rather than to individual users. An administrator registers and tests accounts, discovers models, and monitors usage. Any enabled account intended for general use should therefore have its own provider-side budget and restrictions.

PacketChat has local authentication. It does not require users to maintain their own provider credentials. Email delivery for invitations and password resets can be manual or sent through SMTP or Resend.

What to plan before deployment#

Choose durable locations for PostgreSQL and object storage, generate production secrets, place the web service behind HTTPS, and decide who may administer provider spending. Back up both the database and object store; either one alone is incomplete.

Next#