DocsGPT is an open-source platform for private AI agents, assistants and enterprise search over an organisation's own documents, built by Arc53 and deployable in the cloud, on your own servers or fully air-gapped. Every answer is cited to its source; the licence is MIT.
UK Department for Work and Pensions: an internal knowledge assistant on government infrastructure, with local inference.
Public sector · local inference
AMD
AMD runs DocsGPT as an assistant over internal technical documentation.
Internal technical documentation
LexEU
LexEU: multilingual EU-law research over 2.5M+ pages, answers in 6 EU languages with full source citations.
2.5M+ pages · 6 EU languages · cited
Run it where your data is allowed to be: Cloud · Managed · On-premises · Air-gapped
Same platform, same MIT licence, from one seat. Hybrid (on-premises platform, your cloud LLM) is an On-premises option; air-gapped means zero external network. Or self-host the community edition.
Who operates each posture, where documents are stored and where the language model runs. Cloud and Managed are operated by Arc53; On-premises and Air-gapped run in your own environment with Arc53 support.
Deployment postures: operator, data location, LLM location
Hybrid (on-premises platform + cloud LLM) is listed under On-premises.
One self-hosted RAG platform: knowledge, agents, workflows, governance
Four parts of one open-source codebase.
Knowledge → Agents → Workflows → Governance
01Knowledgeingest · OCR · index
02Agentsretrieve · cite · act
03Workflowsroute · check · approve
04Governanceguardrails · audit · SSO
Text version
Documents are connected, parsed and indexed (Knowledge). Agents answer over them with citations and call tools (Agents). Multi-step jobs run node by node with human approval where a step is uncertain (Workflows). Guardrails, the audit log and SSO/SCIM/RBAC apply across all of it (Governance).
7 guardrail checks that flag, redact or block; human approval for answers and tool calls; append-only audit log; OIDC SSO, SCIM 2.0 and RBAC via Admin. Docs: guardrails
A large Central European commercial bank processes 10,000+ inbound documents a day; time-to-action fell from 24 hours to 16 minutes, with over 99.5 % classification accuracy, running fully air-gapped on the bank's existing NVIDIA T4 GPUs.
Source: EU bank, name under NDA
Government infrastructure, local inference
UK Department for Work and Pensions: an internal knowledge assistant on government infrastructure, with local inference.
Source: DWP
2.5M+ pages · 6 languages
Multilingual EU-law research over 2.5M+ pages, answers in 6 EU languages with full source citations, 100–200 ms retrieval.
Source: LexEU
Built for regulated environments
Identity, audit, guardrails and open code. Threat model and disclosure process on the security page.
Yes. DocsGPT runs on your servers or air-gapped with local models, so no document or question leaves your network; the code is MIT-licensed and inspectable.
Can AI be on prem?
Yes. DocsGPT deploys with Docker Compose or Kubernetes manifests in your data centre and runs local engines such as Ollama, vLLM or llama.cpp; a bank runs it fully air-gapped on existing NVIDIA T4 GPUs.
Is there a self-hosted AI that cites its sources?
Every DocsGPT answer carries title, source and page number, and retrieval can be hybrid (BM25 + vector) or GraphRAG.
What does it cost?
Cloud is $20/seat/month; Managed is from $2,000/month, flat; On-premises and Air-gapped are from $10,000, one-time or recurring — per deployment, not per seat. Individuals can start free in the app. See pricing.
See DocsGPT on your documents
A 30-minute demo on your use case, or start free in the cloud app. Self-hosting takes one command.