Access Control

Access control in Beever Atlas is membership-based: users can only see data from channels they are a member of on the originating platform. This applies to both private AND public channels — a user who has not joined #backend cannot see its memories, even if the channel is public.

Why Membership-Based Access

This design choice prioritizes user expectations over platform defaults:

User expectation: "I should only see data from channels I'm in"

Platform differences:

  • Slack: Public channels are readable by all workspace members
  • Beever Atlas: All channels require membership, regardless of privacy setting

Rationale:

  • Prevents information overload from irrelevant channels
  • Matches user mental model of "my channels"
  • Consistent experience across platforms
  • Simplifies permission model

How Access Control Works

Channel Membership Sync

Beever Atlas pulls channel membership from each platform:

Slack:

members = await slack.conversations_members(channel=channel_id)
info = await slack.conversations_info(channel=channel_id)
is_private = info["channel"]["is_private"]

Discord:

members = await discord.channel_members(channel_id)
is_private = channel.type != ChannelType.text

Microsoft Teams:

members = await teams.channel_members(channel_id)
is_private = channel.visibility != "public"

Storage: MongoDB collection channel_acl

{
  channel_id: "C123456",
  platform: "slack",
  is_private: false,
  member_ids: ["U123", "U456", "U789"],
  last_synced: "2026-04-13T12:00:00Z"
}

Sync frequency:

  • On initial channel sync
  • When user explicitly refreshes
  • Background refresh every 1 hour

Access Checking

Every API request checks access before returning data:

Query request:

# 1. Authenticate user
user_id = await verify_token(request.headers["Authorization"])

# 2. Get accessible channels
accessible = await acl.get_accessible_channels(user_id)

# 3. Filter results
results = await semantic.search(query, channel_ids=accessible)

Wiki request:

# Check if user can access channel
if not await acl.check_access(user_id, channel_id):
    return {"error": "Channel not accessible"}, 403

Filter results:

# Remove results from unauthorized channels
filtered = [r for r in results if r["channel_id"] in accessible]

When channel_id is omitted from a query, Beever Atlas searches across all channels the user is a member of — not all channels in the workspace.

Example:

  • User is member of #backend, #platform, #auth
  • User is NOT member of #frontend, #mobile
  • Query without channel_id → searches #backend, #platform, #auth only
  • Results from #frontend and #mobile are filtered out

Implementation:

# Get user's accessible channels
accessible = await acl.get_accessible_channels(user_id)

# Search across all accessible channels
results = await semantic.search(
    query="deployment",
    channel_ids=accessible  # Only user's channels
)

Platform-Specific Behavior

Slack

Authentication: Bearer token from Slack app installation

Membership:

  • Pulled via conversations.members API
  • Includes private channels user has joined
  • Excludes channels user hasn't joined (even if public)

Bot access: Bot token has access to all channels in workspace

Discord

Authentication: Bot token with server permissions

Membership:

  • Pulled via channel member list
  • Includes all channels user can see
  • Respects Discord role permissions

Bot access: Bot has access to all channels in server

Microsoft Teams

Authentication: Azure AD app with Graph API permissions

Membership:

  • Pulled via Graph API channel members endpoint
  • Respects Teams private channel permissions

Bot access: Bot has access to all teams in tenant

Access Control Integration Points

API Authentication

Bearer token middleware validates user identity before any operation:

@app.middleware("http")
async def authenticate(request: Request, call_next):
    token = request.headers.get("Authorization", "").replace("Bearer ", "")
    if not token:
        return JSONResponse(status_code=401, content={"error": "Missing auth token"})
    
    user = await verify_workspace_token(token)
    request.state.user_id = user.id
    request.state.workspace_id = user.workspace_id
    return await call_next(request)

All routes receive request.state.user_id and request.state.workspace_id after authentication.

Retrieval Pipeline

Semantic Agent and Graph Agent filter results before returning:

async def search_with_acl(query: str, user_id: str):
    # Get user's accessible channels
    accessible = await acl.get_accessible_channels(user_id)
    
    # Search only accessible channels
    results = await weaviate.search(
        query=query,
        filters=[{"channel_id": accessible}]  # Filter by ACL
    )
    
    return results

Wiki Builder

Private channel sections display [restricted] for unauthorized users:

# Engineering Wiki

## Topics

### Authentication ✅
[Full content visible]

### Backend Infrastructure 🔒
[restricted - join #backend to view]

### Frontend 🚫
[restricted - join #frontend to view]

Neo4j Traversal

Global entities (Person, Technology, Project, Team) are visible to all users:

# Alice is visible to all users
alice = await neo4j.get_entity("Alice Chen")

Relationships from private channels are filtered:

# Alice's work on private-project only visible to #private-project members
if "private-project" in user_accessible_channels:
    relationships = await neo4j.get_relationships(alice, "private-project")
else:
    relationships = []  # Filtered out

Security Considerations

Token Validation

Workspace tokens are validated against platform APIs:

  • Slack: auth.test endpoint
  • Discord: Validate against bot token
  • Teams: Validate against Azure AD

Token expiration: Tokens expire per platform policy

  • Slack: User tokens expire after ~12 hours
  • Discord: Bot tokens don't expire
  • Teams: Azure AD tokens expire after 1 hour

Refresh: Clients must refresh tokens before expiration

Membership Freshness

Sync triggers:

  • Manual refresh via API
  • Background refresh every 1 hour
  • On user join/leave events (webhook)

Stale data: If membership sync fails, last known membership is used

  • Conservative: Deny access if sync failed recently (< 5 min ago)
  • Background retry: Sync continues in background

Cross-Workspace Considerations

Workspace isolation: Each workspace has isolated data

  • Slack workspace A cannot access Slack workspace B
  • Discord server A cannot access Discord server B

User identity: User IDs are scoped per workspace

  • Same human in two workspaces = two different user_ids
  • No cross-workspace data leakage

Configuration

# Access Control Settings
ACL_SYNC_INTERVAL_SECONDS=3600  # 1 hour
ACL_STALE_THRESHOLD_SECONDS=300  # 5 minutes

# Authentication
SLACK_SIGNING_SECRET=required   # Slack app verification
DISCORD_PUBLIC_KEY=required     # Discord bot verification
TEAMS_APP_ID=required           # Teams app ID

# Token Validation
TOKEN_VALIDATION_ENABLED=true
TOKEN_CACHE_TTL_SECONDS=300     # 5 minutes

Troubleshooting

User can't see channel they're in

Causes:

  1. ACL sync hasn't run recently
  2. User joined channel after last sync
  3. Sync failed for that channel

Solutions:

  1. Trigger manual ACL refresh: POST /api/admin/acl/refresh
  2. Check sync logs for errors
  3. Verify bot has permissions to read channel members

User sees channel they shouldn't

Causes:

  1. Channel membership sync error
  2. User left channel but ACL not updated
  3. Bot token has excessive permissions

Solutions:

  1. Trigger immediate ACL sync for that channel
  2. Review bot permissions (principle of least privilege)
  3. Audit channel membership in platform vs database

Cross-channel search returns too many/few results

Check:

  1. User's actual channel memberships in platform
  2. ACL data in MongoDB: db.channel_acl.find({member_ids: "USER_ID"})
  3. Sync logs for failed channel syncs

Debug query:

# Get user's accessible channels
accessible = await acl.get_accessible_channels(user_id)
print(f"User {user_id} can access: {accessible}")

# Count memories per channel
counts = await weaviate.count_by_channel(channel_ids=accessible)
print(f"Memories per channel: {counts}")

Next Steps

Agent Architecture

How Google ADK agents orchestrate ingestion, queries, and wiki generation in Beever Atlas

Resilience & Reliability

How Beever Atlas handles failures gracefully and maintains availability when components fail

On this page

Access ControlWhy Membership-Based AccessHow Access Control WorksChannel Membership SyncAccess CheckingCross-Channel SearchPlatform-Specific BehaviorSlackDiscordMicrosoft TeamsAccess Control Integration PointsAPI AuthenticationRetrieval PipelineWiki BuilderNeo4j TraversalSecurity ConsiderationsToken ValidationMembership FreshnessCross-Workspace ConsiderationsConfigurationTroubleshootingUser can't see channel they're inUser sees channel they shouldn'tCross-channel search returns too many/few resultsNext Steps