Workspace memory

By default, every new connection starts cold — your assistant knows nothing about how your team works until you tell it again. Workspace memory fixes that. It gives the concierge (https://hdls.ai/api/mcp) a durable place to store context, so your assistant can pick up where it left off — across sessions, and across the whole team.

These tools live on the concierge endpoint, so they work the moment you're connected — see Getting connected if you haven't set that up yet.

💬 Just ask

"Remember that we sell to mid-market manufacturers and always log a follow-up date."

"What do you know about how my team works?"

You don't call these tools yourself — just tell your assistant; the technical reference below is for when you want the details.

There are two kinds of memory, and the difference matters:

Shared — the workspace profile. One record for the entire workspace. Everyone on the team reads the same thing. "What this workspace is, how we work, conventions the assistant should follow."
Personal — your bio and your preferences. Private to you. Other members of the same workspace cannot see your bio or preferences, even though you all share the same backend.

At a glance

You want to…	Tool	Scope	Who can use it
Read the team's shared profile	`get_workspace`	shared	reader+
Update the team's shared profile	`set_workspace`	shared	admin+
Read your own bio	`get_bio`	personal	member+
Save your own bio	`set_bio`	personal	member+
List your own preferences	`get_preferences`	personal	member+
Save a preference	`set_preference`	personal	member+

Mental model. A workspace is one tenant, and your credential is pinned to it — so you never pass a tenant_id. Personal memory (your bio, your preferences) is private to you even inside a shared workspace. Shared memory (the workspace profile) is visible to everyone on the team.

Shared: the workspace profile

get_workspace returns the team-wide profile — a name, a description, and a freeform data object. It returns null if nothing has been saved yet.

// tool: get_workspace   (reader+)
{}

set_workspace saves or updates it. Because this is shared state that affects everyone, only an admin or owner can change it. Only the fields you pass are touched.

// tool: set_workspace   (admin+)
{
  "name": "Acme Sales",
  "description": "B2B pipeline for the EMEA sales team.",
  "data": {
    "fiscal_year_start": "February",
    "default_currency": "EUR",
    "forecast_cadence": "weekly"
  }
}

Use the profile to record the durable facts you'd otherwise repeat every session: your fiscal calendar, naming conventions, the forecast cadence, "always report in EUR" — anything you want every teammate's assistant to honour automatically.

Personal: your bio

Your bio is who you are in this workspace — stored against your credential and visible only to you. Any member (or above) can read and write their own.

// tool: set_bio   (member+)
{
  "content": "Alex — RevOps lead. I care about clean stage data and an honest forecast.",
  "data": { "timezone": "Australia/Sydney", "focus": "EMEA enterprise deals" }
}

// tool: get_bio   (member+)
{}

A bio is how the assistant remembers who it's working with on a later connect — your role, your timezone, what you focus on — without anyone else on the team seeing it.

Personal: your preferences

Preferences are simple, durable key/value settings the assistant should honour for you later — also private to you.

// tool: set_preference   (member+)
{
  "key": "default_report_format",
  "value": "markdown_table"
}

// tool: get_preferences   (member+)
{}

A few rules:

key must be a simple identifier — letters, digits, and underscores, starting with a letter or underscore (e.g. default_report_format). Keys with spaces, dashes, or punctuation are rejected.
value may be any JSON — a string, number, boolean, array, or object.
Setting the same key again overwrites the previous value.

Preferences are perfect for small standing instructions: a default report format, a preferred currency, "always round amounts to whole dollars."

What persists across sessions

Once written, all three slices survive disconnects and reconnects — they live in the workspace, not in the session. On a fresh connection your assistant can read them back and immediately act in context:

The shared profile so the whole team works the same way.
Your bio so it knows who it's helping.
Your preferences so its output matches how you like to receive it.

That's what turns a stateless tool connection into something that remembers your business.

Gotchas worth knowing

data is replace, not merge. For set_workspace and set_bio, the data object you send fully replaces whatever was stored — it is not a deep merge. To keep existing keys, read first (get_workspace / get_bio), merge locally, then write the complete object back. (Preferences are different — each set_preference touches only its one key.)
Personal stays personal. Your bio and preferences are scoped to your own credential. Another member of the same workspace — even an admin — cannot read them. The shared profile is the only slice the whole team sees.
Shared edits need admin+. Reading the workspace profile is open to readers, but changing it requires admin or owner, because it affects everyone.
Managed workspaces only. Workspace memory lives on hdls-managed workspaces. If your workspace points at your own bring-your-own database, the memory tools aren't available there.

Where to go next

Set up the team that shares this context: Teams & roles.
Pull in ready-made playbooks your preferences can shape: Skills.
Install the products your workspace memory describes: Products.