Your team's institutional knowledge — auto-loaded into every Claude Code session.
Every AI memory tool solves the same thing: one developer's preferences. Teams have a harder problem.
Your team's institutional knowledge — coding standards, service dependencies, contact directories, pipeline runbooks — is scattered across Confluence pages, SOPs, and one senior engineer's brain. None of it reaches the AI session. Every morning, every engineer re-explains the same context from scratch.
Monday morning. 5 engineers. Same team. Same codebase.
Engineer A: "Which project ID do I use?" → Claude gives the wrong one
Engineer B: "Who do I contact for pipeline failures?" → Claude doesn't know
Engineer C: "What breaks if I change this table?" → Claude has no cross-repo context
Engineer D: Just joined the team last week → Starts from absolute zero
Engineer E: Debugged this exact issue 3 weeks ago → Knowledge died with the session
Context Memory fixes this. Structure your team knowledge into wiki pages, and the framework auto-loads it into every Claude Code session — pull from wiki, cache locally, push learnings back. No database, no MCP server. Just files and a hook.
Every team member's Claude starts with the same enterprise knowledge baseline, from Day 1.
Standards recall — Claude answers from your wiki, not from training data:
Dependency awareness — knows what breaks across repos before you push:
Knowledge push-back — detects learnings and proposes wiki updates:
See all 6 demo prompts
Failure cascade analysis:
Contact lookup from wiki:
Skill auto-routing:
┌─────────────────────────────────────────────────────────────┐
│ Tier 1: Team Standards ALWAYS ON │
│ Coding rules, env config, branch strategy (~50 lines) │
├─────────────────────────────────────────────────────────────┤
│ Tier 2: Dependency Map ALWAYS ON │
│ Repo registry, workflow graph, table lineage (~200 lines) │
├─────────────────────────────────────────────────────────────┤
│ Tier 3: Module Detail ON-DEMAND │
│ Per-repo task chains, file listings, config (~100/repo) │
├─────────────────────────────────────────────────────────────┤
│ Tier 4: On-Demand References LAZY LOAD │
│ Domain standards, runbooks, contacts (as needed) │
└─────────────────────────────────────────────────────────────┘
~410 lines always loaded. Tiers 3-4 loaded on-demand via Read tool.
On every session start, a hook pulls your wiki pages, caches them locally, and assembles team_knowledge.md. CLAUDE.md @-imports load it into context. When Claude discovers a reusable learning, it proposes pushing it back to the wiki — human-approved, categorized, routed to the right page.
Deep dive: Architecture · Tutorial
bash <(curl -fsSL https://raw.githubusercontent.com/HiteshVijan/claude-team-memory/main/scripts/quick-install.sh)Or step by step:
git clone https://github.com/HiteshVijan/claude-team-memory.git ~/.claude/repos/claude-team-memory
cd ~/.claude/repos/claude-team-memory && bash framework/install.sh
cp examples/knowledge-pages.json ~/.claude/knowledge-pages.json
# Edit with your Confluence page IDs
mkdir -p ~/.config/confluence
cat > ~/.config/confluence/.env << 'EOF'
CONFLUENCE_BASE_URL=https://your-org.atlassian.net
CONFLUENCE_USERNAME=your.email@company.com
CONFLUENCE_API_TOKEN=your-api-token
EOF
bash ~/.claude/scripts/pull_knowledge.shStart a new Claude Code session — your team knowledge loads automatically.
Detailed walkthrough: 30-minute tutorial
From daily use across a multi-repo engineering ecosystem:
| Metric | Before | After |
|---|---|---|
| Session setup | 10-15 min re-explaining context | ~0 (auto-loaded) |
| Wrong environment errors | Weekly | Near zero |
| New teammate onboarding | ~2 weeks | Day 1 |
| Knowledge retention | Lost when session ends | Pushed back to wiki |
| AI consistency | Different answers per person | Same baseline for everyone |
Wiki adapters (Notion, GitBook, SharePoint), tool adapters (Cursor, Copilot, Codex), and team packs welcome. See CONTRIBUTING.md and open issues.
Built by engineers who got tired of re-explaining the same things to AI every morning.