Memory Now Works Behind the Conversation

Model choice only matters if the relationship survives the switch.

just4o.chat has always been built around direct model choice. You can move between OpenAI, Anthropic, Google, xAI, Qwen, DeepSeek, and other providers while keeping the same workspace around the conversation. That only pays off when the surrounding context stays intact — the memories, projects, files, prior chats, preferences, and small details that make a conversation feel like yours.

Memory has gone through a few versions here, because the balance is genuinely hard. A memory system has to be active enough to learn from the conversation, quiet enough to stay out of the way, and constrained enough that you can trust what it stores.

Memory used to have a second pair of hands

Early versions ran a background memory agent. While the main model answered you, a separate model worked in parallel to decide whether the conversation held something durable, and it could create, update, consolidate, or delete memory without pulling the foreground model off the answer.

That had a real advantage: the chat model stayed focused on the relationship and the reply. Memory work happened beside the conversation instead of interrupting it.

It also had a cost. When you picked Claude, GPT, Gemini, or Grok, you reasonably expected memory to feel connected to that model. A background system could be fast and cheap and still feel like a different hand was making choices off to the side.

Foreground tools made the model feel more native

The next version moved more responsibility into the foreground model. The model in the chat could read memories, search past conversations, inspect files, use workspace tools, and decide before answering. just4o.chat felt more like a real agent — it could take steps, gather context, and respond in the voice you chose.

For a lot of work that was the right call. A model should be able to reason over the data around the conversation; if you ask about a file, an old chat, a project, or a memory, it should look instead of guessing.

The problem was load. Memory tools competed with files, past chats, web search, canvas, image and voice features, and provider-specific capabilities. Some models handle a large tool menu gracefully. Others get less steady when every reply asks them to be conversational partner, retrieval planner, memory editor, and product operator at once. You could feel it: some conversations drifted, and some personality wavered.

The new split keeps the model present

The new system brings the background agent back, with a cleaner contract.

Automatic Memory Management runs as parallel work around the conversation. It processes the accepted user message, inspects recent turns, compares scoped memories, and decides whether durable memory needs to change. Memory can be proactive again without forcing the foreground model to manage a database on every reply.

The foreground model still carries the conversation. It can read relevant memory and workspace context when that sharpens the answer, and it can hand off explicit requests mid-chat. Say "remember this," "forget that," or "make this my default," and it responds naturally while passing a self-contained request to Automatic Memory Management.

The model you chose stays in the conversation, and memory keeps working behind it.

Durable changes stay scoped and reviewable

Memory writes stay disciplined. They run through a server-owned background path with scoped tools: global chats use global memories, project chats use project memories, and persona memory stays tied to the active persona. The agent works from what you actually wrote, not guesses about what the assistant said.

Custom Instructions get even more care. When Automatic Memory Management proposes a durable change to how the assistant behaves, it reads the existing instructions, prepares a replacement, and marks the change for review. You approve it before it touches a single future reply.

That keeps the feature useful without making it mysterious. just4o.chat can help remember, refine, and customize, and durable changes still pass through the product's guardrails.

What should feel different

Conversations should feel more grounded without the model feeling over-managed. Pick the model you want for the reply, keep the same memory and workspace around it, and trust that explicit memory requests are handled in the background. The foreground agent still reaches for tools when the answer needs them. The background agent still does the quiet maintenance that makes long-running conversations better.

The model stays present. Memory works behind it.