Connections
A connection is one of your systems. Its credentials are encrypted and used only on the server - your people connect Claude and use it without ever seeing a key.
New here? How this works in five steps
- Add a connection - pick your system from the list and paste one credential. We handle the rest.
- Tools arrive automatically - templates bring ready-made tools, or import them from a system's API page in one click.
- Make a group - e.g. "Accounting (read-only)" - and tick the tools it should include.
- Add your people to that group.
- They connect Claude and instantly have exactly what you granted - nothing more.
Directory
Staff appear here automatically when they connect Claude. Tick a group in a person's drawer to grant access — it saves instantly, no reconnect, and the confirmation reports exactly what changed and what that person can now reach.
| Person | Status | Groups | Reach | Apps | Keys | Storage |
|---|
Each tool is one thing Claude can do (read = safe lookup, write = changes data). Most are added for you - from a template, or by importing a system's API spec on its connection. Building one by hand is an advanced fallback. Click a row for the full definition, who can reach it, and its actions.
| Tool | Connection | Method & path | Reach |
|---|
Web apps that staff have had Claude build for them - dashboards and mini-pages hosted on Helix at a private URL. Each reads company data with its owner's access through a read-only API key. An app is created, edited, and deleted by its owner from their My apps page; here you can review every app, archive one (it stops serving but keeps everything, restorable any time), and permanently delete an archived app. Click a row for the detail, its files and database, and the actions.
| App | Owner | Status | Data access | Files & size | Updated |
|---|
API keys let a script, cron job, or other non-MCP system reach the library over a plain
REST API (/library/*, described by /openapi.json). Each key
carries a scope (skills, knowledge, tools, audit, actions, or all) and an access level
(read-only or read & write). The secret is shown once at creation - store it safely;
revoke any key here to disable it immediately.
| Key | Kind | Scope | Access | Groups | Last used | Created |
|---|
Analytics
Usage, users, and errors at a glance, so you can see what's working without reading the log line by line. The raw audit log is still available under Audit.
Nothing was recorded in this time window. Try a longer window, or check back once people have used their tools.
Drill-down
Calls over time
Mix
Calls over time
Latency over time (ms)
Top tools by volume
P95 latency by connection (ms)
Calls by connection
Top users by volume
Fault rate by connection
Faults by source
Outcome mix over time
Top faulting tools
Recent errors
View raw audit log →| Time | User | Tool | Status | Took | Source | Reason |
|---|
Apps & API keys
The same window, sliced by hosted app and by API key. Click an app for its own trend, viewers, and recent calls; click a key for the standard drill-down.
App
Calls over time
What it calls
Who used it
Recent calls
| Time | Viewer | Tool | Result | Took | Reason |
|---|
Apps
| App | Owner | Calls | Errors | Refusals | Viewers | P95 |
|---|
API keys
| Key | Kind | Calls | Errors | Refusals | Viewers | P95 |
|---|
Access hygiene
"Unused" only means unused within the audit log's retention window (90 days) — usage older than that has already aged out and cannot be seen here.
| User | Last active | Status | Granted tools | Unused grants |
|---|
Long-term trends
Daily call/user/fault counts, kept in their own small record outside the audit log so they outlast its retention window.
Calls per day
Monthly active users
Storage
Company-wide storage by category, people ranked by how much of their quota is used, and apps ranked by size. To change someone's quota, open their row on the People tab of Directory — quotas are edited there, not here.
Not counted against quotas
Shown for information only — these stores use disk but no one's storage quota: memory notes, knowledge entries, app key-value data, and skills. Sizes and counts only; content is never read.
People by quota use
| Person | Used | Quota | Share used |
|---|
Apps by size
| App | Owner | Size (with backups) | Stored data |
|---|
Groups
A group is a named basket of tools, skills, and actions you hand to people and API keys. Click a row to see everything it grants, who has it, and manage membership. To build a variant of an existing group, open it and use Duplicate - no re-picking from scratch.
| Name | People | Keys | Tools | Skills | Actions | Updated |
|---|
Knowledge (Helix)
Helix is a shared knowledge pool every connected worker can search, read, and contribute to from Claude - a place to capture things people learn. Entries publish immediately; you can edit or remove any of them here. A source connects a document URL (like an llms.txt) and keeps its contents here as a synced entry you refresh on demand. Personal notes belong in each worker's private Memory instead - if one was filed here by mistake, Move to memory hands it back to its author privately.
Memory
Memory is each worker's private note store in Claude - preferences, ongoing work, personal reminders - tied to their account. Content is visible only to its owner, never here: this page shows per-person counts so you can see the feature is being used, nothing more. Shared, company-wide notes live in Knowledge.
| User | Notes | Last updated |
|---|
No one has saved a private note yet. Workers save them from Claude - "remember that I..." - with the memory_add tool.
Skills
Skills are installable bundles (a SKILL.md plus optional files) that workers add to their own Claude. Workers can contribute from Claude; submissions wait here for your approval. Access is granted per group (on the Groups tab), and a skill installs to its scope - project, personal, terminal, or Claude Desktop.
Requests
The backlog of bugs, feedback, tool requests, and feature/change requests about Helix itself, filed by workers from inside Claude - click a row to read a request, add notes, and resolve or decline it.
How reporting works
Workers file requests from Claude with helix_report_bug,
helix_send_feedback, helix_request_access,
helix_request_feature, and helix_request_change. The worker
who filed a request can see your notes and its status from their own portal. A note
that came out wrong can be rewritten or removed - hover it in the thread for
Edit and Delete. An edited note keeps its original
timestamp and is marked as edited; a deleted one is gone for good, for them too. Turn
reporting off to remove all five tools from Claude. You can also author your own
request here with New request; it stays in this queue and is not
shown in any worker's portal.
New request
| Ref | Kind | Sev | Title | Reporter | Notes | Age | Updated | Actions |
|---|
Roadmap
Accepted work, kept separate from the open request queue so "yes, someday" doesn't clutter what's actively being triaged. Add an item two ways: promote a request from its detail page (Add to Roadmap), or author a net-new item below. Work an item in place: move its status (Planned → In progress → Done) and keep notes on its thread - marking it Done posts the outcome back on the request it came from, so the original filer sees what shipped. Removing an item deletes it; it does not reopen any request it came from.
New roadmap item
Actions
Actions are specific commands that run on the Helix server host itself - for wrapping
tools that have no API, or triggering scripts you deploy alongside the server. They are
defined in code (src/actions.ts) and shipped with the server, not created
here; this page lists the catalog. Grant an action to a group on the
Groups tab, exactly like a tool. Each accepts only typed, validated
arguments and never runs through a shell. Turn actions off to remove every one from
Claude and the REST API at once.
Devices
An unattended TV or kiosk display, paired to show exactly one app - no sign-in required
once paired (#118). Add a device to get a 5-digit code, then on the TV go to
/tv and enter it. Whoever's there then picks which app the
display shows - that's permanent for this device; change it here if it needs to switch.
Audit
| Time | User | Tool | Connection | Method | Path | Status | Took | Source | Reason |
|---|
| Time | Key | Viewer | Tool | Connection | Method | Path | Status | Took | Source | Reason |
|---|