---
title: "Overview"
description: "An OpenAI-compatible gateway with persistent memory, personas, and model routing built in."
---

> Documentation Index
> Fetch the complete documentation index at: https://truememory-docs.mvstg.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

The AI Gateway sits between your application and LLM providers. It exposes the same `/v1/chat/completions` endpoint every OpenAI SDK already knows — and layers persistent memory, personas, tool-call routing, and observability on top, so your app doesn't have to build them.

## Base URL and placeholders

Truememory is deployed per environment, so examples across these docs use the placeholder **`https://your-tie-host`** — replace it with your instance's gateway URL (ask whoever operates your deployment; for the Mindvalley staging environment this is provided with your credentials). "TIE" is the platform's internal codename — you'll see it in the placeholder and in a few wire-level names (`tie.bearer`); it's the same product.

Two credentials appear in examples: `$TOKEN` (a user's Bearer token from [Truememory Auth](/authentication/overview/)) and `$AUTH_SECRET` (the backend service-account secret).

## Features

- **OpenAI-compatible** — Drop-in replacement for OpenAI, Cloudflare AI Gateway, or LiteLLM
- **Custom prompt mode** — Send your own `system` or `developer` message to override the agent's built-in prompt while keeping memory and personas
- **Memory** — Persistent user memory via Graphiti knowledge graph, injected automatically
- **Personas** — Custom system prompts per user, injected automatically
- **Client tool calling** — Pass tool definitions per-request, Truememory routes tool_calls back for execution
- **Internal tools** — Memory search/write execute server-side, invisible to the client
- **Thread persistence** — Maintain conversation state across requests
- **Audio** — Speech-to-text and text-to-speech via OpenAI-compatible endpoints
- **Image Generation** — Generate images via OpenAI-compatible `/v1/images/generations` with OpenAI and Google Gemini image models
- **Observability** — All requests traced via Langfuse
- **Multi-agent** — Select different agents via `X-Agent-Id` header

## Supported providers

- [Anthropic](https://docs.anthropic.com/) (via Vertex AI) — Claude Haiku 4.5, Claude Sonnet 4.5, Claude Sonnet 4.6, Claude Opus 4.5, Claude Opus 4.6
- [OpenAI](https://platform.openai.com/) — GPT-5 Nano, GPT-5 Mini, GPT-5.1
- [Google Vertex AI](https://cloud.google.com/vertex-ai) — Gemini 2.5 Flash, Gemini 2.5 Flash Lite, Gemini 2.5 Pro, Gemini 3 Flash (preview), Gemini 3.1 Pro (preview)
- [Z.AI](https://docs.z.ai/) — the GLM family (GLM-4.x, GLM-5.x)

Query `GET /info` on your instance to see which models are currently available — it returns the exact wire IDs. Browse them with runnable examples in the [model catalog](/models/).

### Model ID formats

The `model` value is provider-prefixed, and the prefix depends on the endpoint:

| Endpoint | Anthropic | OpenAI | Google | Z.AI |
|---|---|---|---|---|
| `/v1/chat/completions` | `anthropic/claude-sonnet-4-6` | `openai/gpt-5.1` | `vertexai/gemini-2.5-flash` | `zai/glm-4.7` |
| `/v1/audio/*`, `/v1/images/*` | — | bare IDs accepted (`whisper-1`, `gpt-4o-mini-tts`) | `google/gemini-2.5-flash-tts`, `google/gemini-2.5-flash-image` | — |

Yes, Google appears as `vertexai/` on chat and `google/` on media endpoints — copy the prefix from the endpoint's own docs when in doubt.

## Where to start

| Page | Description |
|------|-------------|
| [Unified API](/ai-gateway/chat-completions/) | Send messages via the OpenAI-compatible `/v1/chat/completions` endpoint |
| [Agents](/ai-gateway/agents/) | Multi-agent routing, agent selection, and the `X-Agent-Id` header |
| [Memory](/memory/overview/) | Persistent memory, context prefetch, and memory management |
| [Personas](/ai-gateway/personas/) | Custom system prompts per user — create, activate, and manage |

Source: https://truememory-docs.mvstg.com/ai-gateway/overview/index.mdx