DocsGPT is an open-source private AI assistant and self-hosted RAG (Retrieval-Augmented Generation) platform developed by Arc53, providing private AI agents, assistants, and enterprise search capabilities for an organization's own documents. The platform is licensed under MIT, all answers cite their sources, and it can be deployed in the cloud, on your own servers, or in fully air-gapped environments.
Purpose
DocsGPT is used to build private AI assistants, agents, and enterprise search on top of an organization's internal documents, enabling organizations to run AI where their data permits. Every answer cites its sources.
Key Capabilities
The platform consists of four parts built from the same open-source codebase:
Knowledge: Connect, parse, and index documents. Supports 30+ file formats, provides connectors such as Google Drive, SharePoint, Confluence, GitHub, and S3, and supports OCR for scanned PDFs, wikis, and knowledge graphs.
Agents: Agents answer questions based on indexed documents and provide citations, and can call tools. Supports Classic, Agentic, Research, and Workflow agent types, supports scheduled tasks, connected tools, and any MCP server, and comes with its own MCP server so that Claude or Cursor can search its knowledge base.
Workflows: A visual builder that lets you drag and drop Start, AI Agent, Set State, If / Else, Code, Note, and End nodes onto a canvas; multi-step tasks run node by node, with human approval supported at uncertain steps; every run keeps a step-by-step trace.
Governance: 7 guardrail checks that can flag, redact, or block content on top of a company-wide baseline; human approval for answers and tool calls; support for roles, usage quotas, exportable audit logs, and run traces; OIDC SSO and SCIM 2.0 supported through Admin.
How to Use
Supported on macOS / Linux and Windows, requires Docker, and the Community Edition can be self-hosted with a single command. A cloud-hosted version is also available at app.docsgpt.cloud. Documentation provides a quickstart guide.
Deployment Modes
- Cloud (Cloud): app.docsgpt.cloud, a multi-tenant environment operated by Arc53, suitable for teams without data residency restrictions.
- Managed: a dedicated instance, operated by Arc53, with no need to run it yourself.
- On-premises: deployed in the customer's data center, supported by Arc53, suitable for strict data residency requirements; hybrid mode (on-premises platform + cloud LLM) is an on-premises deployment option.
- Air-gapped: deployed in the customer's environment, with zero external network, suitable for zero data egress requirements, supported by Arc53.
Use cases
Designed for regulated environments, suitable for organizations with requirements for identity authentication, auditing, guardrails, and open-source code. Typical scenarios include:
- Government departments: the UK Department for Work and Pensions (DWP) runs an internal knowledge assistant on government infrastructure, using on-premises inference.
- Enterprise internal technical documentation: AMD uses DocsGPT as an internal technical documentation assistant.
- Multilingual legal research: LexEU conducts multilingual EU legal research based on more than 2.5 million pages of content, answers in 6 EU languages, and provides complete source citations.
- Financial industry: a large Central European commercial bank processes more than 10,000 inbound documents daily, reducing action time from 24 hours to 16 minutes, with classification accuracy exceeding 99.5%, running fully air-gapped on the bank's existing NVIDIA T4 GPUs (source: EU bank, name protected by NDA).
Limitations
No other explicit limitation information is available at this time. The materials do not provide further explanation regarding specific missing features, performance ceilings, or unsuitable scenarios.
Pricing
Cloud is billed per seat at $20 per person per month; other deployment modes are priced per deployment rather than per seat. Managed dedicated instances start at $2,000 per month (fixed price). On-premises and air-gapped deployments are one-time or recurring fees, billed per deployment, with zero data egress in air-gapped mode.
