Skip to main content

​
Memory & RAG

The 60db CLI exposes the full Memory/RAG layer for scripting and automation. Every billable command surfaces the wallet balance and charge in the output footer so you can watch spend in real time.
All memory commands are pay-as-you-go. See the pricing reference for rates. When the wallet runs out, commands return INSUFFICIENT_CREDITS with a structured shortfall — top up via the Dashboard billing page to continue.

​
Commands

CommandBilledPurpose
60db memory:ingest✓Store a single memory
60db memory:upload✓✓Extract + ingest a document (PDF, DOCX, XLSX, images…)
60db memory:search✓Hybrid semantic search
60db memory:context✓Assemble LLM-ready context
60db memory:collections—List collections
60db memory:create-collection—Create team/knowledge/hive collection
60db memory:usage—Monthly spend breakdown
60db memory:status—Poll a memory’s ingestion status
60db memory:delete—Soft-delete a memory (24h undo)

​
Ingest a memory

Store a single fact, preference, or text snippet: 
60db memory:ingest \
  --text "User prefers dark mode and metric units" \
  --type user \
  --title "UI preferences"
Options:
  • -t, --text <text> — Memory content (required, max 100,000 chars)
  • --title <title> — Optional display title
  • -c, --collection <id> — Target collection (defaults to personal)
  • --type <type> — user | knowledge | hive (default: user)
  • --no-infer — Skip LLM-based fact extraction (default: infer is on)
Cost: $0.0001 per 1,000 characters.

​
Upload a document

Upload a file — PDF, DOCX, XLSX, PPTX, EML, MSG, HTML, TXT, Markdown, CSV, PNG, JPG, TIFF, and 70+ other formats — with built-in OCR for scanned content: 
60db memory:upload \
  --file ~/Documents/quarterly-report.pdf \
  --collection company_reports \
  --type knowledge \
  --title "Q4 2026 Report"
Options:
  • -f, --file <path> — Absolute or relative path to the document (required)
  • -c, --collection <id> — Target collection
  • --type <type> — user | knowledge | hive (default: knowledge)
  • --title <title> — Display title (defaults to filename)
  • --chunk-size <n> — Max characters per chunk, 200-8000 (default: 1500)
  • --chunk-overlap <n> — Character overlap between chunks (default: 200)
Max file size: 200 MB. Max chunks per document: 100. Cost (two-stage):
  • Extract fee: $0.003 per MB
  • Ingest fee: $0.0001 per 1,000 extracted characters
Both fees are automatically refunded on any failure.

​
Tuning tips

Document type--chunk-size--chunk-overlap
Technical docs / API refs1500200 (default)
Long-form prose / books2500300
FAQs / short snippets800100
Spreadsheet exports30000
Scanned PDFs (OCR)2000250

​
Search memories

Hybrid semantic + keyword recall with optional cross-encoder reranking: 
60db memory:search \
  --query "What is the refund policy?" \
  --mode fast \
  --limit 5 \
  --alpha 0.8
Options:
  • -q, --query <text> — Search query (required, max 2,000 chars)
  • -c, --collection <id> — Collection to search
  • --mode <mode> — fast (~100-200ms) or thinking (wider candidate pool + cross-encoder rerank, ~200-400ms)
  • --limit <n> — Max results (1-50, default: 10)
  • --alpha <n> — 0 (keyword only) to 1 (semantic only), default 0.8
  • --recency-bias <n> — Weight for newer memories (0-1, default: 0)
  • --graph — Include knowledge-graph relationships
Advanced reranker knobs (override server defaults):
  • --rerank-top-k <n> — Max candidates the cross-encoder reranks
  • --rerank-timeout <ms> — Hard timeout for rerank call
  • --min-rerank-score <n> — Drop results below this rerank score (0-1)
  • --fetch-multiplier <n> — In thinking mode, fetch N x limit candidates
Cost: flat $0.0003 per query. When the cross-encoder reranker is active, results include both a dense score (vector similarity) and a rerank score (cross-encoder confidence). The rerank score is the more reliable signal for ranking quality.

​
Example — tune by query type

# Exact phrase match (keyword-heavy)
60db memory:search -q "invoice #INV-2026-042" --alpha 0.2

# Conceptual question (semantic-heavy)
60db memory:search -q "how do I configure SSO?" --alpha 0.9

# Complex multi-faceted question (thinking mode + reranking)
60db memory:search -q "what are common escalation patterns?" --mode thinking --alpha 0.7

# Recent events focus
60db memory:search -q "product updates" --alpha 0.6 --recency-bias 0.3

# Override reranker settings for this query
60db memory:search -q "detailed compliance requirements" --mode thinking --min-rerank-score 0.3 --fetch-multiplier 5

​
Assemble LLM context

Purpose-built for RAG — returns a pre-formatted string ready to prepend to an LLM system message: 
60db memory:context \
  --query "Tell me about customer preferences" \
  --top-k 8 \
  --max-context-length 2000
Options:
  • -q, --query <text> — Query driving retrieval (required)
  • --session-id <id> — Chat session ID for hierarchical context
  • --top-k <n> — Max memories to pull (default: 10)
  • --max-context-length <n> — Max assembled tokens (default: 4000)
  • --graph — Include graph relationships
  • --no-timeline — Exclude recent events
Graceful degradation: returns an empty prompt on outage — chat keeps working. Cost: flat $0.0005 per query.

​
Manage collections

​
List collections

60db memory:collections
Shows every collection visible to you in the current workspace with kind (personal/team/knowledge/hive) and shared status.

​
Create a collection

Admin/owner only. Personal collections are auto-created per user — you don’t create those manually. 
60db memory:create-collection \
  --id customer_support \
  --label "Customer Support KB" \
  --kind knowledge
Options:
  • -i, --id <id> — Collection ID (required, lowercase alphanumeric + underscores)
  • -l, --label <label> — Human-readable label (required)
  • -k, --kind <kind> — team | knowledge | hive (default: team)
  • --no-shared — Don’t share with workspace members
Both commands are unbilled.

​
Monitor spend

60db memory:usage
60db memory:usage --period last_30_days
60db memory:usage --period all_time
Shows net spend, operation count, refund count, and a per-service-type breakdown (ingest / extract / recall / context). Always free. Works even when the wallet is empty. Options:
  • --period <period> — current_month (default) | last_30_days | all_time

​
Poll a memory’s status

Memories are processed asynchronously. Use this to watch a specific ID: 
60db memory:status --id mem_01HV8K2X3N4P5Q6R7S8T9U
Statuses: pending → processing → ready (or failed). Unbilled.

​
Delete a memory

60db memory:delete --id mem_01HV8K2X3N4P5Q6R7S8T9U
Soft-delete with 24-hour undo grace period. You can delete memories you created; admins/owners can delete any memory.

​
Agent-friendly JSON output

All memory commands respect the global --json flag for machine-readable output: 
60db --json memory:search --query "user preferences" --limit 5
Returns the same data as the API but with a structured envelope that includes a billing: {charged, balance, txId} object on billable commands.

​
Handling insufficient credits

When the wallet balance is below the operation cost, commands return a 402-style error with structured details: 
$ 60db memory:ingest --text "hello" --type user
✗ Insufficient credits in workspace wallet
  required: $0.000100  available: $0.000010  shortfall: $0.000090
ℹ Top up your workspace wallet via the Dashboard: https://app.60db.ai
The --json output surfaces the same details as: 
{
  "success": false,
  "error": "INSUFFICIENT_CREDITS",
  "details": {
    "required": 0.0001,
    "available": 0.00001,
    "shortfall": 0.00009
  }
}