Wiki Generation
Beever Atlas automatically transforms your team conversations into a persistent, browsable wiki. Unlike traditional wikis that require manual editing, Beever Atlas wiki pages are generated from your structured memory and kept up-to-date as conversations continue.
Why Auto-Generated Wiki Matters
Traditional wikis suffer from three problems:
- Manual maintenance burden: Someone must write and update pages
- Stale content: Wikis quickly become outdated as no one updates them
- Incomplete coverage: Only formally documented knowledge makes it in
Beever Atlas solves all three:
- Zero maintenance: Wiki pages auto-generate from conversations
- Always up-to-date: Pages update when new messages arrive
- Complete coverage: Everything discussed is captured
Wiki Hierarchy
The wiki structure mirrors the 3-tier semantic memory hierarchy:
Level 1: Channel Overview
High-level summary of the entire channel:
Source: Tier 0 semantic memory (channel summary)
Content:
- What this channel is about
- Recent activity overview
- Key themes and topics
- Important decisions or changes
Update frequency: After each channel sync
Example:
# #engineering Overview
The engineering team discusses architecture decisions,
deployment strategies, and technical implementation.
**Recent activity**:
- Decided to migrate from OAuth to JWT for authentication
- Planning database migration for Q2
- Evaluating Terraform for infrastructure
**Key themes**: Authentication, Deployment, InfrastructureLevel 2: Topic Pages
Grouped knowledge about specific themes:
Source: Tier 1 semantic memory (topic clusters)
Content:
- Topic summary
- Key facts (5-10 most relevant)
- Related decisions
- People involved
- Related projects
Update frequency: Incremental after sync, full rebuild daily
Example:
# Authentication
## Summary
The team uses JWT with RS256 for API authentication.
Migrated from OAuth in January 2026.
## Key Facts
- Decision: Switch from OAuth to JWT for stateless auth (Jan 15)
- Implementation: Alice assigned to implement JWT (Jan 16)
- Security: RS256 algorithm chosen for key rotation support (Jan 17)
## Related Decisions
- [JWT Authentication Decision](/decisions/jwt-auth)
## People Involved
- [Alice Chen](/people/alice-chen) - Implementation lead
- [Bob Smith](/people/bob-smith) - Security review
## Related Projects
- [API v2 Migration](/projects/api-v2)Level 3: Entity Pages
Dedicated pages for people, decisions, projects, and technologies:
Source: Neo4j graph memory
Content:
- Entity details
- Related entities
- Temporal evolution
- Source conversations
Update frequency: After entity extraction changes
Example (Person):
# Alice Chen
**Role**: Senior Software Engineer
**Team**: Backend Platform
## Projects
- [API v2 Migration](/projects/api-v2) - Lead
- [JWT Authentication](/decisions/jwt-auth) - Implementation
## Recent Decisions
- Decided to use JWT with RS256 (Jan 15)
- Proposed Terraform for infrastructure (Jan 10)
## Relationships
- Works on: API v2 Migration
- Decided: JWT Authentication
- Member of: Backend Platform teamExample (Decision):
# JWT Authentication Decision
**Status**: ✅ Implemented
**Date**: January 15, 2026
**Decided by**: Alice Chen
## Summary
Replace OAuth with JWT for stateless authentication.
Uses RS256 for key rotation support.
## Rationale
- OAuth requires session store, JWT is stateless
- RS256 allows non-repudiation and key rotation
- Simpler integration with microservices
## Evolution
- **Jan 15**: Initial decision to use JWT
- **Jan 17**: Confirmed RS256 algorithm (supersedes HS256 proposal)
- **Feb 1**: Implementation completed
## Related Entities
- **Blocks**: Refresh token rotation (pending)
- **Supersedes**: OAuth Authentication Decision
- **Used by**: API v2 MigrationGeneration Triggers
Wiki pages generate on three triggers:
1. After Sync (Incremental)
When: Automatically after each channel sync completes
What:
- Assign new facts to existing or new clusters
- Update touched cluster summaries
- Update channel summary
- Mark wiki as dirty
Cost: ~$0.01 per 100 new messages
Latency: ~10 seconds for typical sync
2. Daily Full Rebuild
When: 2 AM UTC every day
What:
- Re-evaluate all clusters for coherence
- Split clusters > 100 members
- Merge clusters with similarity > 0.85
- Rebuild cluster summaries
- Full wiki regeneration
Purpose: Ensures long-term quality and coherence
Cost: ~$0.50 per channel with 10,000 messages
3. On-Demand (Manual)
When: Triggered via API by admin
What:
- Force full cluster reconsolidation
- Rebuild entire wiki
- Update all entity pages
Use cases:
- After bulk import of historical messages
- After changing clustering parameters
- To fix quality issues
API:
POST /api/admin/wiki/rebuild
{
"channel_id": "C123456",
"force": true
}Wiki Components
Summaries
Generated by LLM from clustered facts:
Channel Summary:
- Input: All cluster summaries
- Output: High-level channel overview
- Length: 2-3 paragraphs
Cluster Summary:
- Input: Cluster member facts
- Output: Topic overview
- Length: 1-2 paragraphs
Cost: ~$0.01 per summary
Key Facts Selection
Top 5-10 facts per cluster, ranked by:
- Quality score: Higher-quality facts preferred
- Temporal decay: Recent facts boosted (except decisions)
- Citation count: Frequently cited facts boosted
- Importance: High/critical importance always included
Exemptions from decay:
- Decisions, architecture, policies (slow decay)
- High/critical importance (no decay)
- Frequently cited (reinforced)
Cross-References
Automatic linking between related content:
Topic → Decisions:
- Decision entities mentioned in topic facts
- Links to decision pages
Topic → People:
- People entities mentioned in topic facts
- Links to person pages
Decision → Projects:
- Projects affected by decision
- Links to project pages
Entity → Facts:
- Source conversations mentioning entity
- Links to relevant message clusters
Browsing the Wiki
Web Dashboard
The web UI provides wiki browsing:
Navigation:
- Channel list → Select channel → Browse wiki
- Topic tree → Expand topics → View facts
- Entity index → Browse people, decisions, projects
Search:
- Full-text search across wiki content
- Filter by topic, entity type, date range
- Semantic search ("find decisions about auth")
Citations:
- Every fact links to source message
- Jump directly to Slack/Discord/Teams
- View full conversation context
API Access
Wiki content available via REST API:
# Get channel wiki
GET /api/wiki/:channel_id
# Get topic page
GET /api/wiki/:channel_id/topics/:topic_id
# Get entity page
GET /api/wiki/entities/:entity_type/:entity_idWiki Quality
Cluster Health Rules
Applied during daily full rebuild:
| Condition | Action | Rationale |
|---|---|---|
| Cluster > 100 members | Split via k-means | Too broad, needs subdivision |
| Similarity > 0.85 | Merge clusters | Duplicates or near-duplicates |
| Coherence < 0.4 | Re-cluster | Lost coherence over time |
| 0 members | Delete cluster | Empty cluster |
Quality Metrics
Cluster Coherence:
- Average pairwise similarity of members
- Target: > 0.4
- Below threshold → re-cluster from scratch
Summary Quality:
- LLM-rated for relevance and completeness
- Poor summaries regenerated
- Target: > 0.7 quality score
Coverage:
- Percentage of facts assigned to clusters
- Target: > 95%
- Unclustered facts flagged for review
Cost Optimization
The wiki is designed for cost efficiency:
Read Costs:
- Cached wiki pages = FREE
- No LLM calls to serve cached content
- Cache hit rate: > 80% for typical traffic
Write Costs:
- Incremental updates: ~$0.01 per 100 messages
- Daily rebuild: ~$0.50 per 10K messages
- Amortized cost: ~$0.001 per message per day
Optimization:
- Dirty flag prevents unnecessary regeneration
- Differential updates only touch changed clusters
- Batch processing for large syncs
Comparison to Traditional Wikis
| Aspect | Traditional Wiki (Confluence/Notion) | Beever Atlas Wiki |
|---|---|---|
| Content creation | Manual writing | Auto-generated from chat |
| Maintenance | High (manual updates) | Zero (auto-updates) |
| Freshness | Stales quickly | Always up-to-date |
| Coverage | Limited to documented topics | Everything discussed |
| Consistency | Varies by author | Consistent LLM generation |
| Cost | Editor time | ~$0.001/message/day |
| Temporal tracking | Manual page history | Automatic evolution chains |
| Relationships | Manual links | Auto-generated from graph |
Next Steps
- See the Ingestion Pipeline that populates the wiki
- Learn about the Dual Memory Architecture that enables wiki generation
- Understand Query Router that uses the wiki for fast answers