Every tool holds its own key
Each AI tool asks for its own OAuth grant to Gmail, Drive or the CRM. Eight tools and six sources is 48 grants, each reviewed, rotated and revoked separately.
Early access for enterprise teams
One governed layer between your company's data and every AI tool.
Connect Google Workspace, your CRM, GitHub and internal databases once. Every AI tool gets only what each person may see.
What is Helios?
Helios is a governed access layer between a company's data sources and the AI tools its people use. Each source is connected once. Claude, ChatGPT, Cursor and in-house agents connect to Helios over MCP (Model Context Protocol) or REST and act as the signed-in person, receiving only the records and fields that person's role allows. Every call is audited.
Last updated
Why do AI tools need a governed layer?
Every AI tool your company adopts asks for the same thing on day one: a key to your data. The keys multiply faster than anyone can review them.
No tool knows your org chart
A grant opens everything the account can see. Nothing stops a recruiting agent from reading payroll threads or a sales copilot from pulling HR notes.
Source APIs pay for the sprawl
Every tool polls the same mailboxes, calendars and pipelines on its own schedule, and Google and HubSpot start answering with rate-limit errors.
How does Helios work?
Four steps, done once per company. After that, adding an AI tool takes one URL and a sign-in.
Connect each source once
People sign in with their company account and approve each source a single time. Helios stores every grant encrypted and never hands it to an AI tool.
Bring the policies you already have
Roles come from your Google Workspace directory and Jamf. Start from eight role templates, adjust them, and assign them by dragging. Admin changes always win over automatic ones.
Connect any AI tool by URL
Add one MCP address to Claude, ChatGPT, Cursor or your own agent and sign in with OAuth. Agents that cannot speak MCP use the REST API or a bootstrap link.
Every answer filtered and audited
Each record is passed, masked or blocked according to the person's role before it leaves Helios. Every call is written to one audit trail with the person, agent and account.
What does each AI agent actually see?
Every role policy has an access level. Each class of data, from contact details to deal values, is passed, masked or blocked at each level before it leaves Helios. Pick an agent to compare.
Whose agent is asking
HubSpot contact
as returned to the support rep's agent
- Name
- Dana W.
- Masked
- d•••@northwind.com
- Masked
- Phone
- +1 ••• ••• 0142
- Masked
- Deal value
- $100k to $250k
- Masked
- Notes
- Asked for a payment plan
- Passed
What's in Helios
Built for data that lives in SaaS tools and belongs to individual people.
Acts only as the signed-in person
No service accounts, bot identities or shared keys. An agent can never see more than the person it works for.
Content screened by meaning
Payslips, medical matters and personal threads are blocked. One-on-ones and HR discussion arrive tagged private. If screening is unavailable, nothing is relayed.
Policies from your org chart
Titles and departments from the Google directory and Jamf groups map people to role policies automatically, daily and for every new joiner.
Three switches on every tool
A tool runs only when the role policy, the person's own setting and the calling token all allow it. Each is set by a different person for a different reason.
hubspot.search_deals runs
Pass, mask or block each field
Contact details, deal values, account numbers and IDs are passed, masked or blocked by access level before data leaves Helios.
One read, many subscribers
A source is read once and served to every agent allowed to see it, so load on Google, HubSpot and internal APIs grows with sources, not agents.
Events, not polling
Agents subscribe to new mail, calendar changes, shared files, chat mentions, pull requests and failed builds, delivered by signed webhook or queue. The event sources are set up for each deployment.
Discovery that scales
Agents call find_server, then find_tools for that server. About 500 tokens of tool definitions, however many sources you connect.
How do agents connect?
Over MCP (Model Context Protocol) or REST. Agents see four tools: find_server, find_tools, call_tool and call_write_tool. That is about 500 tokens of definitions, however many sources you connect.
# Connect once, from any MCP client
$ claude mcp add --transport http helios https://helios.yourcompany.com/mcp
# 1. Find the server that can answer
find_server { "query": "deals closing this quarter" }
→ [{ "id": "srv_7f2", "name": "hubspot-sales" }, { "id": "srv_c40", "name": "finance-dw" }]
# 2. Ask that server for tools
find_tools { "server_id": "srv_7f2", "query": "deals by close date" }
→ [{ "name": "hubspot.search_deals" }, { "name": "hubspot.get_deal" }]
# 3. Run it, as the signed-in person
call_tool { "name": "hubspot.search_deals", "arguments": { "close": "this_quarter" } }
→ { "deals": 14, "masked": ["amount"], "blocked": { "not_relevant": 2 } }What will your security team ask?
The short answers. We share the full architecture during discovery.
- Identity
- MCP clients sign in with OAuth 2.1 and PKCE. Access tokens are ES256-signed and last one hour. Revoking a connection stops it on its next request.
- Credentials
- Source grants are encrypted at rest and never leave Helios. Agent tokens are stored only as SHA-256 hashes and shown once.
- Data handling
- Screening and masking happen before data leaves Helios. When the screening service cannot decide, the result is blocked rather than relayed.
- Audit
- Every call records the person, the agent or app, the account that acted, the tool, masked arguments, duration and outcome. Filter by any of them.
Helios vs. connecting each AI tool directly
Credentials
Connecting each AI tool directly: Every tool holds its own grant to every source
Through Helios: Each source connected once; tools never hold a grant
Access policy
Connecting each AI tool directly: Whatever the account can see
Through Helios: Role policies from your directory, enforced on every call
Sensitive fields
Connecting each AI tool directly: Returned in full
Through Helios: Passed, masked or blocked by access level
Audit
Connecting each AI tool directly: Scattered across vendors, if it exists
Through Helios: One trail: person, agent, account, tool and result
Load on source APIs
Connecting each AI tool directly: Grows with every tool
Through Helios: One read serves every allowed subscriber
Adding an AI tool
Connecting each AI tool directly: New grants, new review, new policy work
Through Helios: One URL and a sign-in
Revoking access
Connecting each AI tool directly: Per tool, per source
Through Helios: One switch, effective on the next request
| Aspect | Connecting each AI tool directly | Through Helios |
|---|---|---|
| Credentials | Every tool holds its own grant to every source | Each source connected once; tools never hold a grant |
| Access policy | Whatever the account can see | Role policies from your directory, enforced on every call |
| Sensitive fields | Returned in full | Passed, masked or blocked by access level |
| Audit | Scattered across vendors, if it exists | One trail: person, agent, account, tool and result |
| Load on source APIs | Grows with every tool | One read serves every allowed subscriber |
| Adding an AI tool | New grants, new review, new policy work | One URL and a sign-in |
| Revoking access | Per tool, per source | One switch, effective on the next request |
Frequently asked questions
What is Helios?
Helios is a governed access layer between a company's data sources and the AI tools its people use. Each source is connected once. Claude, ChatGPT, Cursor and in-house agents connect to Helios over MCP (Model Context Protocol) or REST and act as the signed-in person, receiving only the records and fields that person's role allows. Every call is audited.
How is Helios different from an MCP gateway?
An MCP gateway routes agent calls to services. Helios also decides what each person may see: it acts as the signed-in person, applies role policies imported from your directory, masks or blocks sensitive fields, and audits every call. It is built for data that lives in SaaS tools and belongs to individual people.
Which data sources does Helios connect to?
Helios connects to the sources each company uses. Helios already connects Google Workspace (Gmail, Calendar, Drive, Chat, contacts and the directory), HubSpot CRM and GitHub. Each enterprise deployment is set up for that company's own sources, such as other CRMs, marketing platforms and internal databases.
Which AI tools work with Helios?
Any MCP client that supports OAuth sign-in, including Claude, ChatGPT and Cursor, plus your own agents. Agents that cannot use MCP can call the REST API or follow a bootstrap link.
How does field-level masking work?
Every role policy has an access level, and every class of data has an action at each level: pass, mask or block. A support rep's agent might see a contact as "Dana W." with a masked phone number, while the account owner's agent sees the full record and an external partner's agent sees nothing.
Who decides what each AI tool can see?
Admins assign role policies, which can be imported from the Google directory and Jamf. Each person can switch tools off for the agents acting for them, and each connection can be narrowed further. A tool runs only when all three allow it, and admin decisions always override automatic ones.
Does Helios reduce load on Google, HubSpot and internal APIs?
Yes. Helios reads each source once and serves the result to every agent allowed to see it, so the number of calls grows with the number of sources rather than the number of AI tools.
How do we get access?
Helios is in early access for enterprise teams. Register interest with the sources and AI tools you want to connect first, and the Techjays team will arrange a discovery call to scope your deployment: which sources and AI tools to connect, and whether Helios runs on your premises or as a managed service.
Register interest
We're onboarding a small number of enterprise teams. Tell us what you would connect first, and we'll come back with a deployment plan for your sources and AI tools.
- A deployment plan for the sources and AI tools you name
- The full architecture for your security review
- A reply from the Techjays team at your work email