Cass Memory

Installation

# Install from GitHub (npm)
npm install -g Dicklesworthstone/cass_memory_system

Health Commands

# Version check
cm --version

# Quick health check
cm quickstart --json

# Full diagnostics
cm doctor --json

# Repair degraded local setup
cm doctor --fix --no-interactive

Usage Patterns

Local Memory (Default)

Session logs remain local by default.

# Get task-specific context before non-trivial DX work
cm context "repair MCP daemon EOF issue" --json

# Inspect the actual context payload
cm context "prefer z.ai coding endpoints first for this coding lane" --json | jq '.data.relevantBullets'

# Store a memory/playbook bullet
cm playbook add "Always use worktrees for canonical repos" --category workflow

# Inspect rules
cm playbook list

# Broader similarity search when context is too narrow
cm similar "worktree recovery" --threshold 0.1 --json

Cross-Agent Sharing (Opt-In)

Cross-agent sharing uses sanitized summaries only.

# Enable sharing
export CASS_SHARE_MEMORY=1

# Disable sharing
export CASS_NO_SHARE=1

Contract

1. Local-first: Session logs remain local by default
2. Opt-in sharing: Cross-agent sharing must be explicitly enabled
3. Sanitized output: Never persist raw secrets, raw transcripts, or tokens
4. No IDE config: CLI-native, not rendered to IDE MCP configs
5. Primary agent read path: use cm context "<task>" --json before trying broader retrieval
6. Candidate-first writes: agent judgment can nominate candidates, but durable shared memory requires promotion
7. Task-shaped retrieval: phrase reads like an actual operator task; use cm similar for loose wording checks
8. Cross-VM split: upstream remoteCass is for SSH-based remote history, while promoted-rule sharing should use explicit playbook export/import

Controls

For the current upstream CLI, explicit privacy controls are also available via:

cm privacy status
cm privacy enable
cm privacy disable

Expected Output

• Local playbook updates
• Optional redacted digest records for fleet learning
• Decision logs for recurring patterns

Fleet Sync Integration

CASS Memory is managed by Fleet Sync as a CLI tool:

# Check health via Fleet Sync
~/agent-skills/scripts/dx-mcp-tools-sync.sh --check --json | jq '.tools[] | select(.tool=="cass-memory")'

# Install via Fleet Sync
~/agent-skills/scripts/dx-mcp-tools-sync.sh --apply --json

Upstream

• Repo: https://github.com/Dicklesworthstone/cass_memory_system
• Docs: https://github.com/Dicklesworthstone/cass_memory_system#readme

Validation

Layer 1 (Host Runtime)

cm --version && cm quickstart --json && cm doctor --json

Layer 4 (Client Visibility)

• NOT REQUIRED - CLI tools don't appear in codex mcp list or similar

Related

• fleet-sync: Fleet Sync orchestrator
• llm-tldr: MCP static analysis
• serena: MCP symbol-aware edits + memory