Sana Agents Review

Turn meetings and connected company knowledge into answers, agents, and approved actions.

Independently researched by AI Toolbox Team · Reviewed 2026-07-15
THE SHORT VERSION

What Sana Agents does

Sana Agents is a knowledge assistant that connects meetings and business sources to cited chat, reusable prompts, integrations, and multi-step AI agents.

Sana Agents combines company knowledge, meeting capture, chat, prompt templates, integrations, and configurable agents. Official tier documentation lists meeting connections for Google Meet, Teams, and Zoom plus knowledge integrations such as Google Drive, SharePoint, OneDrive, Notion, and Confluence. Team and enterprise deployments expand document scale, models, administration, custom integrations, APIs, and analytics. A useful rollout starts with a bounded question set and reliable sources, not unrestricted access to every file and external tool.

The official comparison reviewed July 15, 2026 describes Free as up to five workspace members with ten meetings and twenty messages per user monthly, while Team supports up to fifty members, broader integrations, ten thousand documents per integration, unlimited queries and recordings, and model choice. Enterprise adds SAML SSO, SCIM, domain verification, DPA, additional models, API or custom integrations, analytics, SLA, and support. The source does not state a Team price, so organizations must request and verify current commercial terms, limits, storage, implementation, and overages.

Sana explicitly states that no AI system can be completely protected from prompt injection. Its described defense layers include model detection, permission boundaries, admin guardrails, trusted domains, integration scoping, and human approval for write or external actions. Those controls still require testing. Meeting audio can contain secrets or malicious instructions, citations can point to outdated content, and summaries can invent owners or decisions. Obtain participant consent, minimize recordings and data access, review citations, and require the accountable person to approve messages, transactions, decisions, or record changes.

UNDER THE HOOD

How Sana Agents works

A workspace connects approved calendars, meetings, document systems, and optional tools. Sana indexes permitted knowledge for chat and configurable agents, which retrieve context, invoke an available language model, cite sources, and may plan multi-step work. Platform boundaries, trusted domains, and explicit human approval can constrain writes or external actions.

YOUR INPUTSANA AGENTSREVIEWED OUTPUT
QUICK START

How to set up Sana Agents

1

Define a bounded use case

Document sources, owners, participants, meeting consent, retention, allowed domains, prohibited tasks, success measures, and the person accountable for final actions.

2

Secure the workspace

Configure identity and lifecycle controls, restrict memberships and models, connect only approved folders and calendars, and limit MCP servers, integrations, and external domains.

3

Build narrow agents

Use precise instructions, named sources, structured outputs, error handling, and approval gates; avoid broad write scopes or unattended access to consequential systems.

4

Run adversarial evaluation

Test hallucination, stale or conflicting evidence, private content, prompt injection inside documents and meetings, blocked domains, tool misuse, and permission changes.

5

Pilot and monitor

Train users to inspect citations, review recordings and summaries, monitor usage and actions, revoke stale access, investigate failures, and pause unsafe agents quickly.

COMMON QUESTIONS

Sana Agents FAQs

Is Sana Agents free?

A limited Free tier is documented with per-user meeting and message allowances. Team and Enterprise add scale and controls; verify current pricing directly.

Which knowledge sources can it connect?

Official examples include Google Drive, SharePoint, OneDrive, Notion, and Confluence. Available integrations and limits vary by tier and can change.

Can agents resist every prompt-injection attack?

No. Sana explicitly says complete protection is impossible and recommends layered detection, platform permissions, admin guardrails, and human approval.

Does Sana train models on customer data?

Sana states it does not train underlying language models on customer data. Buyers should verify subprocessors, models, deployment region, retention, and contract terms.

Should meeting summaries be treated as records?

Only after participant review. Transcription and generation can miss context or invent decisions, owners, dates, and commitments.

Listing reviewed 2026-07-15. Product details and pricing can change; verify important terms on the provider's website.

KEEP RESEARCHING

Related Meetings AI tools

Related AI guides

COMMUNITY NOTES

Reviews

Be the first to share a detailed review.

Tell the community what you made, what worked, and what you wish you knew before starting.