Sana
The most open enterprise knowledge agent we have scored: real model choice, 100+ connectors, an MCP client and a genuinely usable free tier. It is also now a Workday subsidiary with no MCP server of its own, no eval surface, and a shared-integration permission model that can show people files they cannot open in SharePoint.
PRICING
| Sana Agents Free | $0 | 10 meetings and 20 chat messages per user per month, up to 5 members, unlimited agents and prompt templates, meeting integrations (Meet, Teams, Zoom), knowledge integrations (Drive, SharePoint, OneDrive, Notion, Confluence), 1,000 documents per integration |
| Sana Agents Team | $30/user/mo | Up to 50 members, unlimited queries and recordings, all off-the-shelf integrations, OpenAI and Claude model selection, 10,000 documents per integration, priority support |
| Sana Agents Civic | $0 | Swedish AI Reform tier: 30 meetings and 200 chat messages per month for Swedish municipalities, agencies, schools, universities, research institutes and non-profits. Application required |
| Sana Agents Enterprise | Quote only | Unlimited members and documents, SAML SSO, SCIM, domain verification, extended range of LLMs, analytics dashboard, MCP client, single-tenant or VPC deployment, dedicated success team and SLA |
| Sana Learn Core | $13/licence/mo | Minimum 300 licences, so about $46,800 per year floor. AI authoring, content library, AI tutor, analytics, automated learning management |
| Sana Learn Enterprise | Quote only | Adds SSO, SCIM provisioning, open API access, HRIS connectors and SLA-backed support |
Sana is unusually transparent for an enterprise vendor: Agents Free and Team are published with real numbers, and the Learn pricing page gives an indicative per-user figure by volume band. Euro figures are converted from published USD at roughly 0.92. The trap is the Learn side: 300 licences is a floor, not a starting point, so a 150-person company effectively pays double the headline rate.
checked 2026-08-07 · vendor pricing page
Element scores
Strengths
Sana does the thing almost no enterprise AI vendor does: it lets you choose the model. Team tier selects between OpenAI and Claude, Enterprise opens an extended range of providers, and DeepSeek R1 was pushed out for Enterprise early testing so agents could reason before answering. Default models are not trained on customer content and Sana states it runs Zero-Day Retention arrangements with third parties where possible. Around that sits the strongest context layer in this category: 100+ connectors re-synced every 24 hours, per-query source scoping, full-page visual PDF analysis including charts and tables, MP4 and Office and CSV handling, meeting capture across Zoom, Teams and Meet, and a 50+ language assistant. Connectors are not read-only wrappers either; the GitHub connector alone exposes 28 tools including create pull request and update file contents, with admin-level allowlisting per tool. Add an MCP client for anything not natively supported, a no-code agent builder, Sheets for batch processing, and a public API in beta, and you have a platform that can actually finish work rather than summarise it. Security posture is solid: ISO 27001, SOC 2 Type 2, GDPR, SSO, SCIM, RBAC, AES-256 at rest, TLS 1.2+ in transit, single-tenant or VPC deployment on Enterprise. And the free tier is real, not a demo: unlimited agents and prompt templates, five integrations, 1,000 documents.
Honest dings
Start with the permission model, because it is the one that can hurt you. Sana's own documentation says shared integrations do not auto-mirror source permissions, that visibility depends on who can reach the shared integration inside Sana, and that file-level restriction is not supported. Their stated remedy is to keep the audience narrow and remove sites from scope if someone sees something in Sana they cannot see in SharePoint. That is a configuration burden sitting on top of a knowledge tool an entire company queries. Third-party MCP servers carry an explicit disclaimer too: Sana cannot verify their quality or security and the customer is responsible. Second gap: no evals. A feedback thumbs and an Enterprise impact dashboard are not a way to measure whether an agent answered correctly, and there is no regression testing or red-teaming surface. Third: protocols run one way. Sana is an excellent MCP client and not an MCP server, so you cannot pull Sana knowledge into Claude, Copilot or your own agent. Compare that to Outreach, a far less open platform overall, which ships 41 MCP tools outward. Then the ownership question. Workday closed the $1.1 billion acquisition in November 2025 and both companies now position Sana Enterprise as the AI operating system layered on Workday's data model, with Workday as the control plane for permissions and process. Sana Labs has confirmed Learn continues as a standalone product, but anyone buying Sana today is buying into a Workday roadmap. Finally, Learn's 300-licence floor prices out any organisation under 300 people, and competitors point out the AI-native design lacks the structured progression and tracking that mandatory compliance training needs.
Sources
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- newsroom.workday.com/2025-09-16-Workday-Signs-Defin… — Official Workday press release 16/09/2025: $1.1B acquisition, Sana founded 2016, Sana Learn and Sana Agents served 1M+ users across hundreds of enterprises, both products to continue, target reach of 75M Workday users (accessed 2026-08-07)
- finsmes.com/2025/11/workday-completes-acquisition-o… — Independent confirmation that the acquisition actually closed in November 2025, ahead of the guided Q4 FY2026 window (accessed 2026-08-07)
- futurumgroup.com/insights/is-workdays-1-1b-sana-acq… — Independent analyst assessment (Keith Kirkpatrick, Futurum, 18/09/2025): Workday spent nearly 2% of market cap, Sana follows Paradox, Flowise, Evisort and HiredScore in an AI acquisition run, Elliott Management disclosed a 4% stake the same day (accessed 2026-08-07)
- sanalabs.com/agent-pricing — Official pricing: Free $0, Team $30/user/month, Enterprise custom. Confirms LLM agnosticism, Zero-Day Retention with third parties, single-tenant deployment, AES-256 at rest and TLS 1.2+ in transit, ISO 27001 and SOC 2, and that Enterprise gets the MCP client and analytics dashboard (accessed 2026-08-07)
- support.sana.ai/en/articles/157324-sana-agents-free… — Official tier comparison (11/12/2025): Free capped at 10 meetings and 20 chat messages per user per month and 5 members, Team adds OpenAI and Claude model selection and 10k documents per integration, Enterprise adds additional LLM models and the analytics dashboard, plus the Swedish AI Reform Civic tier (accessed 2026-08-07)
- support.sana.ai/en/articles/357502-mcp-integration-… — Official MCP docs and the source of two key limitations: 'Sana is not able to verify the quality or security of servers you choose to install', no support for MCP servers requiring manual OAuth endpoint configuration, and admins cannot preview available tools before setup. Confirms streamable HTTP with SSE fallback and backend-only requests (accessed 2026-08-07)
- support.sana.ai/en/articles/148543-integrations-101 — Official, and the source of the permission finding: private integrations mirror user access, shared integrations do not ('Source permissions are not auto-mirrored in Sana'), file-level restriction is not supported, all integration files re-sync every 24 hours (accessed 2026-08-07)
- support.sana.ai/en/articles/662994-github — Official connector documentation used as the function-calling evidence sample: 28 named tools on one connector including create pull request, create or update file contents, create repository and toggle workflow, with OAuth scopes enumerated (accessed 2026-08-07)
- support.sana.ai/en/articles/361684-create-content-a… — Official Summer Release notes: document and slide generation with export to PDF/Docs/Word/PowerPoint, data analysis translated into code, model picker in the chat box, public API in beta for chat, tools and integrations (Enterprise), visual PDF page analysis, enterprise data retention policies, MCP connector launch (accessed 2026-08-07)
- support.sana.ai/en/articles/148586-sana-ai-101 — Official product guide: semantic search over concepts not keywords, source selection, web-search toggle, History activity log, no training of language models on customer data, ISO 27001 / SOC 2 / GDPR (accessed 2026-08-07)
- support.sana.ai/en/articles/269307-deepseek-r1-new-… — Official: DeepSeek R1 made available for early testing on Sana Agents Enterprise so agents can apply deep logic before responding. Primary evidence for the reasoning-model score (accessed 2026-08-07)
- sanalabs.com/products/sana — Official product page: model-agnostic positioning, 100+ integrations, security feature list including audit logging and regional deploys, iOS app, and the customer outcome claims (62% prep time saved at a global law firm, valuation memo from a week to three or four hours) (accessed 2026-08-07)
- sanalabs.com/pricing — Official Sana Learn pricing page: per-user monthly pricing varies by volume band and HQ country, plans start at 300 users, all-in-one platform with AI authoring, AI tutor, analytics and SSO (accessed 2026-08-07)
- cybernews.com/ai-tools/sana-ai-review — Independent review scoring both products 4.2/5, corroborating $30/user/month for Agents and $13/licence with a 300-licence minimum for Learn, and confirming no training on customer data with ISO 27001 and SOC 2 (accessed 2026-08-07)
- courseplatformsreview.com/blog/sana-learn-review — Independent review (updated July 2026): Core plan floor of $46,800 per year, a 150-person company effectively pays $26/user/month against the 300-seat minimum, implementation typically 2-4 weeks, Enterprise adds SSO, SCIM, open API and HRIS connectors. Confirms Sana Labs will continue Learn as a standalone product post-acquisition (accessed 2026-08-07)
- continu.com/compare/continu-vs-sana — Competitor-published comparison, used only for the specific structural criticism: the AI-native design lacks the structured progression, consistent delivery and tracking that mandatory and compliance training require, and the 300-user minimum excludes smaller organisations (accessed 2026-08-07)
- sanalabs.com/agents-blog/ai-tools-supercharge-busin… — Vendor blog, used for the 2026 positioning statement: Sana Core as the AI-native Workday UI assumed for all Workday customers, Sana Enterprise as the cross-tool AI operating system, with Workday as the control plane for permissions and process (accessed 2026-08-07)