Introducing Atlas 1.0, workspace intelligence for fast moving teams

Most tools treat lost context as a personal failure. You should have taken better notes, written a cleaner summary, checked in more often. In practice, context disappears because work is fast, distributed, and spread across too many surfaces for any one person to hold at once.

That is not a productivity problem. It is a structural one, and it does not have a personal solution.

Atlas 1.0 is our attempt at a structural answer.

What Atlas was before today

When we first shipped Atlas, the model was simple: you ask, it responds. You opened the Atlas panel, prompted it, and it would reason over whatever documents you had open. Useful. Bounded.

The ceiling was the session. Atlas knew what you showed it. Nothing more.

What we actually wanted was a workspace that paid attention so you did not have to.

That required a different architecture entirely. Not a smarter model. A different relationship between the AI and the state of your work.

The context graph

At the core of Atlas 1.0 is a continuously updated semantic graph of everything that lives in your workspace. Projects, documents, decisions, threads, and the relationships between them are indexed as the work happens, not on demand.

When you open a project, Atlas has already traversed that graph. It knows what changed since you were last here, what is unresolved, and what connects to work happening in other parts of the workspace. You do not trigger a fetch. The graph is live.

The indexing runs incrementally. Large workspaces do not produce a noticeable lag. We have been running Atlas on workspaces with several hundred active documents and the surface time is consistently under two seconds.

The shift: from assistant to intelligence

Atlas 1.0 maintains a live representation of your workspace. Not a snapshot. A continuously updated graph of your projects, documents, decisions, and their relationships, indexed semantically and kept current as work moves.

You do not summon this. It is already there.

Here is how Atlas resolves context before the model ever sees your prompt:

# Atlas context resolution
context:
  source: /atlas/narrative
  depth: full
  references:
    - project: launch-narrative
      sections: ["risks", "timeline", "open-questions"]
    - thread: friday-review-prep
      status: unresolved
    - document: q1-retrospective
      relevance_score: 0.91

prompt: >
  Draft a 5-bullet status update for Friday's review.
  Surface open risks. Flag what needs a decision before EOW.

# Atlas context resolution
context:
  source: /atlas/narrative
  depth: full
  references:
    - project: launch-narrative
      sections: ["risks", "timeline", "open-questions"]
    - thread: friday-review-prep
      status: unresolved
    - document: q1-retrospective
      relevance_score: 0.91

prompt: >
  Draft a 5-bullet status update for Friday's review.
  Surface open risks. Flag what needs a decision before EOW.

# Atlas context resolution
context:
  source: /atlas/narrative
  depth: full
  references:
    - project: launch-narrative
      sections: ["risks", "timeline", "open-questions"]
    - thread: friday-review-prep
      status: unresolved
    - document: q1-retrospective
      relevance_score: 0.91

prompt: >
  Draft a 5-bullet status update for Friday's review.
  Surface open risks. Flag what needs a decision before EOW.

The resolved context is assembled before the model runs. It never prompts you for more information because it already has it.

Three things Atlas 1.0 can now do that it could not before

1. Proactive surfaces

Atlas flags things without being asked. A project running behind its last stated timeline. A decision from six weeks ago that a current document appears to contradict. A thread that three different team members have referenced without resolution. These surfaces appear in the Atlas panel when you open a workspace, not buried in a digest. They are actionable, and each one has a draft response attached.

2. Cross-project reasoning

Atlas now reasons across your entire workspace, not just within a single document. If two projects are moving toward a conflict, Atlas sees it. If a decision made in one context has downstream implications for a project the author was not part of, Atlas connects them. Single-document context is easy. Organisation-level context is the hard problem, and it is what v1 could not touch.

3. Drafting with full workspace context

When you ask Atlas to draft something, you do not paste in context or open the relevant documents first. You write the prompt. The rest is automatic.

curl https://api.slate.app/v1/atlas/query \
  --header "Authorization: Bearer $SLATE_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "workspace_id": "ws_northwind",
    "context_depth": "full",
    "query": "What are the unresolved risks across active projects this week?",
    "format": "bullets",
    "include_sources": true
  }'

curl https://api.slate.app/v1/atlas/query \
  --header "Authorization: Bearer $SLATE_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "workspace_id": "ws_northwind",
    "context_depth": "full",
    "query": "What are the unresolved risks across active projects this week?",
    "format": "bullets",
    "include_sources": true
  }'

curl https://api.slate.app/v1/atlas/query \
  --header "Authorization: Bearer $SLATE_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "workspace_id": "ws_northwind",
    "context_depth": "full",
    "query": "What are the unresolved risks across active projects this week?",
    "format": "bullets",
    "include_sources": true
  }'

The response includes a sources array alongside the answer. Every surface Atlas produces, proactive or on-demand, is attributed. You can trace every claim back to the document it came from.

Configuring Atlas for your workspace

Atlas behavior is configurable per workspace. The defaults are designed to be useful immediately, but teams with specific privacy requirements or narrower context preferences can adjust them.

atlas:
  context_depth: full           # minimal | standard | full
  proactive_surfaces: true
  cross_project_reasoning: true
  source_attribution: always    # always | on_request | never
  index_scope:
    include: ["projects", "documents", "threads", "decisions"]
    exclude: ["archived", "personal-drafts"]
  retention:
    context_graph_days: 90

atlas:
  context_depth: full           # minimal | standard | full
  proactive_surfaces: true
  cross_project_reasoning: true
  source_attribution: always    # always | on_request | never
  index_scope:
    include: ["projects", "documents", "threads", "decisions"]
    exclude: ["archived", "personal-drafts"]
  retention:
    context_graph_days: 90

atlas:
  context_depth: full           # minimal | standard | full
  proactive_surfaces: true
  cross_project_reasoning: true
  source_attribution: always    # always | on_request | never
  index_scope:
    include: ["projects", "documents", "threads", "decisions"]
    exclude: ["archived", "personal-drafts"]
  retention:
    context_graph_days: 90

The index_scope field lets you exclude specific document types or folders from Atlas indexing entirely. Content in those paths is invisible to Atlas regardless of what is asked.

What does not change

Atlas does not have access to anything outside Slate. No email, no external web access, no data from integrations unless you have explicitly connected a source through the integrations panel.

This was a deliberate choice. Broad ambient access sounds powerful in a demo. In practice it produces noise, not clarity. The value of Atlas is precision, knowing your work deeply rather than knowing everything loosely.

Plan availability

Atlas 1.0 is available across all plans today, with meaningful differences in depth:

The Atlas API is available on Team. Full reference in the documentation.

Why this took longer than we planned

Context-aware AI is not a new idea. The hard part is not the model. It is building the indexing layer, the context graph, the resolution logic that makes the output reflect the real state of your work rather than a plausible-sounding approximation of it. That work is quiet and easy to rush in ways that only surface months later.

We ran Atlas internally for the better part of three months before shipping it. The signal we were waiting for was not that it always got things right. It was that when it got things wrong, the error was visible and correctable rather than subtle and misleading. There is a meaningful difference between those two failure modes. The second erodes trust faster than no AI at all.

Atlas 1.0 is live in all workspaces today. If you are on Free and want access to cross-workspace intelligence, the Team plan upgrade is straightforward.

We are genuinely curious what you build with it.