Skip to content
Overview
Architecture
Comparison
Docs
Sign in
Connect to the API
Overview
Architecture
Comparison
Docs
Sign in
Mirabase App
Connect to the API
Architecture
04
Context compiler
Fit a token budget
Draw on files, knowledge, and memory
Cite every item
03
Long-term memory
Distill facts from conversations
Replace old values; keep history
Confirm, correct, or forget facts
02
Knowledge base
Index every write automatically
Hybrid search with citations
Flag duplicates, superseded content, and conflicts
01
File system
Read and write by path
Version every write; roll back
Branch and snapshot
Architecture
Write
Memory updates
Retrieval
Answer
←
1 / 7
→
Play
CALLER
MIRABASE API
MODELS
SPACE
FILE SYSTEM
KNOWLEDGE BASE
MEMORY
WORKER
driven by a job queue
STORAGE
Request path
Background
Agent / SDK / MCP
PUT …/files
Auth and rate limits
API key, capabilities,
read and write roots
Write
concurrency control,
idempotency, quota
Retrieval
full-text search,
vector search, reranking
Context pack
budget by source
Answer
citations per sentence
Embedding
Reranking
Answering and
distillation
Directory tree
read and write
by path
Versions
history, diff,
revert
Branches and
snapshots
merge, as-of read
Chunks
Vectors
semantic search
Full text
keyword search
Tables
lookup, aggregates
Whole-Entry
index
Freshness buffer
instant search
Facts
with source quotes
Revision links
replaced_by
MEMORY.md
generated
read-only view
Parsing
documents, PDF, spreadsheets,
images, recordings, web pages
Chunking and embedding
Publishing
full-text index,
vector index
Memory distillation
facts with source quotes,
candidate or confirmed
Attribute revision
new values replace old ones
Offline consolidation
quote checks, dormancy,
merge proposals
Main database
source of truth for Entries, versions,
permissions, full text, and facts
Vector index
derived from the main database, rebuildable
Object storage
original files