Skip to content
Start free

DocsGPT vs Onyx

Onyx and DocsGPT are both open-source, self-hostable platforms for chat and search over company data. Onyx's strengths are connector breadth, community size and enterprise logos; DocsGPT's are agent workflows with deterministic steps, OCR for complex documents, built-in evaluation and benchmarking, delivery references in regulated environments — and single sign-on, SCIM, audit export and quotas in the MIT code rather than an Enterprise tier.

Book a demoSelf-host in 5 minutes

Last updated: · Facts checked on

Onyx and DocsGPT at a glance

Licence, hosting, models, price, contract and data access, side by side.

Onyx is MIT at the core with a proprietary ee/ directory. The core self-hosts free, including air-gapped; the quote-based Enterprise tier adds single sign-on and vendor-supported on-premises deployment. Model support covers cloud and local, and Onyx Cloud publishes a price of about $20 per user per month on annual billing. DocsGPT is MIT throughout (platform and CLI), runs in four postures including air-gapped from one seat, supports any provider or local engine, and publishes its Cloud price with flat pricing for self-hosted deployments.

Onyx and DocsGPT compared on 7 rows
 OnyxDocsGPT
Open-source licenceMIT core + proprietary ee/1MIT (platform and CLI)2
Self-host and air-gapAir-gap yes; self-host yes, free (MIT core); vendor-supported on-premises deployment is an Enterprise-tier offer1Yes — Self-hosted (community), On-premises and Air-gapped (zero external egress), from one seat2
Model choice / BYOMCloud and local providers1Any cloud provider or local engine (Ollama, vLLM, llama.cpp and others); BYOM2
Transparent pricingCloud ~$20/user/month on annual billing, published; Enterprise tier quote-based1Yes, per posture: Cloud $20/seat/month; Managed from $2,000/month, flat; On-premises and Air-gapped from $10,000, one-time or recurring2
Seat minimum / contractNone: the MIT core can be self-hosted; no Cloud seat floor recorded1None; self-hosting has no licence fee2
Vendor access to dataYes on Onyx Cloud (Onyx operates the service); No when self-hosted1Yes on Cloud and Managed — Arc53 operates both, Managed as a dedicated instance in your region; No on On-premises and Air-gapped, which you operate2
Enterprise features in the open-source codeQuote-based Enterprise tier: SSO, on-premises and region-specific deployment, white-label, SLA1In the MIT code, free: OIDC single sign-on, SCIM, roles, an exportable audit log, usage quotas, run traces, an MCP server and MCP tools, backup and restore2

¹ Onyx facts checked on . ·² Sources: docs.docsgpt.cloud and the public repository.

Where the products overlap

Both projects cover the same core; the differences are in what sits around it.

Onyx and DocsGPT both offer chat over company documents with citations, connectors to common sources, retrieval-augmented generation over a vector index, and free self-hosting of an MIT-licensed core. Both also sell a hosted cloud plan. Single sign-on is not part of the shared ground: in Onyx it is an Enterprise-tier add-on. Both have the same shape — ingest, index, ask, cite; the sections below list where each is the stronger fit.

DocsGPT GitHub stars

18,317

on GitHub

Contributors

239+

Licence

MIT

platform and CLI

Where Onyx is the better fit

  • Connector breadth: 40+ connectors, covering a wide SaaS estate.
  • A larger community by GitHub stars and contributors.
  • Enterprise references including Netflix, Ramp and Thales.
  • You want a widely adopted open-source search product first, agents second.

Where DocsGPT is the better fit

  • A workflow builder that mixes deterministic steps, LLM steps and human approval.
  • An OCR pipeline for scanned and complex documents with tables.
  • docsgpt-cli bench evaluation suites for regression-safe model swaps.
  • Delivery references in regulated environments: a bank running air-gapped on existing T4 GPUs, the UK Department for Work and Pensions, LexEU.
  • One MIT licence for the platform and the CLI, enterprise controls included: OIDC SSO, SCIM, roles, an audit log exported as CSV or NDJSON, usage quotas, run traces and backup and restore — no paid tier for single sign-on.
  • MCP both ways: agents call any MCP server as a tool, and assistants such as Claude or Cursor search an agent's knowledge through its MCP server.
Workflow builder
  1. startStartchat · api · webhook · schedule
  2. ai agentClassifysmall model
  3. ai agentExtractwrites state
  4. if / elseDecideon shared state
  5. endNotifytemplate
humanHuman approvalapprove · reject · logged
Text version

1. Start (chat · api · webhook · schedule) → 2. Classify (small model) → 3. Extract (writes state) → 4. Decide (on shared state) → 5. Notify (template). Branch from Decide: else · confidence < 0.8 → Human approval → Notify.

zsh — docsgpt-cli bench

24 h → 16 min

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

Cost at 100 and 1,000 seats

List price × seats × 12, before negotiation. DocsGPT Cloud is per seat; Managed and On-premises are flat per deployment, from the published starting prices.

Annual cost at 100 and 1,000 seats
 Onyx CloudOnyx self-hostedDocsGPT
100 seats, per year$24,000/yrFree core; ee/ features licensed separatelyCloud $24,000/yr · Managed from $24,000/yr · On-premises from $10,000, one-time or recurring
1,000 seats, per year$240,000/yrFree core; ee/ features licensed separatelyCloud $240,000/yr · Managed from $24,000/yr · On-premises from $10,000, one-time or recurring
Basis~$20/user/month published, annual billing × seats × 12no licence fee for the MIT core; the ee/ licence terms are not published; not a per-seat figureCloud $20/seat/month × seats × 12 · Managed from $2,000/month, flat · On-premises and Air-gapped from $10,000, one-time or recurring, plus your infrastructure

Before negotiation; add-ons excluded; checked on 2026-09-06.

Annual cost vs seatslist prices · 2026-09-06
Annual cost versus seats: Onyx per-seat list price against DocsGPT Cloud per seat and a flat Managed instanceOnyx annual cost rises linearly with seats from zero to $240,000 at 1000 seats. DocsGPT Cloud rises to $240,000; DocsGPT Managed is flat at $24,000 from its from-price.$0$60k$120k$180k$240k02505007501,000SEATSOnyx, $20/user/mo listDocsGPT Cloud, $20/seat/moDocsGPT Managed, from $2,000/month flat ($24,000/yr)
Text version

At 100 seats Onyx costs $24,000 per year at the $20 per user per month list price; at 1,000 seats, $240,000. DocsGPT Cloud at $20 per seat per month costs $24,000 per year at 100 seats and $240,000 at 1,000; a Managed instance is flat from $24,000 per year at any seat count; On-premises and Air-gapped start from $10,000, one-time or recurring, plus your own infrastructure. Facts checked on 2026-09-06.

Frequently asked questions

Is Onyx open source?

Onyx's core is MIT-licensed with a proprietary ee/ directory for enterprise features (checked 2026-09-06). DocsGPT is MIT-licensed for the whole platform and the CLI.

Is Onyx the same as Danswer?

Yes. Onyx is the current name of that project; the repository, cloud product and Enterprise tier all continue under the Onyx name.

What is the difference between Onyx and DocsGPT?

Both are open-source, self-hostable platforms for chat and search over company data with citations and connectors. Onyx's strengths are connector breadth and community size; DocsGPT adds a workflow builder with deterministic steps and human approval, an OCR pipeline for complex documents, and built-in benchmarking, with delivery references in regulated environments.

Can DocsGPT run air-gapped?

Yes. The Air-gapped posture runs the platform, embeddings and a local LLM engine inside your network with zero external egress, from one seat. A large Central European bank runs it this way on existing NVIDIA T4 GPUs.

Alex Tushynski

Co-Founder, Arc53

Spotted an error? Request a correction and we will fix it and update the date.

See DocsGPT on your documents

A 30-minute demo on your use case, or a free on-premises review with an engineer.

Book a demoSelf-host in 5 minutes