Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/TrinaxCode/TrinaxAI/llms.txt

Use this file to discover all available pages before exploring further.

TrinaxAI provides two complementary ways to give your AI assistant long-term context: Memory for facts about you and your preferences, and Collections for organising indexed code and documents into named, queryable namespaces.

​
Memory

Memory lets you tell TrinaxAI things you want it to remember across conversations. When you say “remember that we use PostgreSQL for production databases” or “remember I prefer TypeScript over JavaScript”, TrinaxAI stores that fact locally and injects it into future prompts.

​
How Memory Persists

Memory entries are stored in storage/user_memory.json as a list of records, each with a unique id, the text of the fact, a created_at Unix timestamp, and optional tags. This file lives entirely on your machine. When you trigger a summary refresh, the LLM reads all stored memories and synthesises them into a concise 3–5 sentence summary that gets injected into the system prompt for subsequent chats. The summary is saved to storage/user_memory_summary.json. The MemoryPanel component in the PWA also supports free-form project notes stored in localStorage under the tc-project-memory key — a scratchpad that doesn’t go through the API.

​
Memory API Endpoints

MethodEndpointDescription
GET/v1/memoryList all stored memory entries
POST/v1/memoryAdd a new memory entry (text, optional tags)
DELETE/v1/memory/{id}Remove a specific memory entry by ID
POST/v1/memory/refreshSummarise all memories into a short context note via LLM
GET/v1/memory/summaryRead the current LLM-generated memory summary
All memory endpoints require localhost or a valid admin token (same as other system endpoints).

​
Memory CLI

# Open the interactive memory manager
trinaxai memory

# The CLI lists entries, lets you add new ones, and delete by ID

​
Memory in the PWA

Open the Memory panel from the PWA sidebar or Settings. The panel shows:
  • Project Notes — a free-form text area stored locally in the browser
  • Auto Summary — the LLM-generated summary of your stored facts, with a Refresh button to regenerate it
  • Add Memory — an input field for text and optional comma-separated tags
  • Memory List — all stored entries with timestamps, tags, and a delete button
When you edit or add memories, the PWA dispatches a tc-memory-updated event so the ChatInterface drops its cached summary and picks up the new version on the next request.

​
Collections

Collections are separate RAG namespaces within the same vector store. Each indexed chunk carries a collection_id metadata field. When you query with a specific collection active, only chunks from that collection are retrieved — keeping project A’s code from contaminating answers about project B.

​
Why Use Collections

Per-Project Isolation

Index each client project or repository into its own collection. Ask questions about “my-app” without getting answers from “legacy-api”.

Mixed Queries

Pass multiple collection_id values in the collections param to query several at once — useful for monorepos or related projects.

Separate Lifecycles

Delete a collection when a project ends. Deletion removes all associated vector nodes from the store and cleans up the manifest entries.

Named Contexts

Give collections meaningful names like “Backend API”, “Design System”, or “Client Docs”. The name appears in source citations in the PWA.

​
Creating and Managing Collections

# List all collections
trinaxai collections

# Index into a named collection
TRINAXAI_COLLECTION_ID=backend \
TRINAXAI_COLLECTION_NAME="Backend API" \
trinaxai index ~/projects/backend-api

# Or set in .env:
# TRINAXAI_COLLECTION_ID=backend
# TRINAXAI_COLLECTION_NAME=Backend API

​
Querying Multiple Collections

Pass an array of collection IDs in the collections field of the chat request to retrieve across multiple namespaces simultaneously: 
curl -X POST https://localhost:3333/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role": "user", "content": "How does auth work?"}],
    "stream": false,
    "collections": ["backend", "auth-service"]
  }'
When collections is omitted or empty, TrinaxAI retrieves from all collections.

​
Default Collection

The default collection (id: "default", name: "General") always exists and cannot be deleted. Files indexed without specifying a collection ID go into default.

​
Cross-Device Sync with App State

TrinaxAI uses an app_state.json file (in storage/) as a simple key-value store for shared state across devices on the same WiFi. All keys use the tc- prefix. The /app-state endpoint lets the PWA on your phone and the PWA on your desktop stay in sync: 
# Read shared state
GET /app-state

# Write/merge keys
PUT /app-state   body: {"values": {"tc-my-key": "value"}}

# Factory-reset state (requires confirmation header)
DELETE /app-state
This is how collection selections, onboarding completion status, and other preferences propagate to other devices without a cloud service.

​
Configuration Reference

VariableDefaultDescription
TRINAXAI_COLLECTION_IDdefaultCollection to use when indexing or querying
TRINAXAI_COLLECTION_NAMEGeneralDisplay name for the active collection

Build docs developers (and LLMs) love

Get started for freeTalk to us