-
Notifications
You must be signed in to change notification settings - Fork 833
-
Hi all,
I'm wondering whether I should follow the Brain / frontal-lobe retirement changes (commit 0a10b3f) from the main repo in my fork, and I ran into a real architecture question.
I did read the commit message, and I understand the core rationale: frontal-lobe mainly existed to preserve continuity across compacted chat sessions, and with workspaces becoming the main substrate, that continuity is expected to come instead from the workspace transcript / scrollback / working tree / local markdown files, rather than from one global persistent self-note.
What I’m unsure about is how far that reasoning should be taken in a setup where frontal-lobe is part of the live monitoring and decision loop, rather than just a continuity aid for chat.
In my fork, frontal-lobe is currently doing more than just "memory" work: it carries regime context, temporary no-trade constraints, open hypotheses, and general continuity across market-monitor, heartbeat, daily-review, chat, and some operator/UI flows.
So my assumption is that removing it naively would reduce hidden prompt state and make syncing changes easier, but it might also make trading behavior less consistent unless there’s a clear replacement for the useful carry-forward state it currently provides.
Can anyone here pls help me understand:
- why the team decided to remove
Brain/frontal-lobein practice beyond the commit summary - whether the intended replacement is simply workspace-local files and transcript continuity, or something more structured
- what you’d recommend as the right way to preserve the useful functionality it can provide today
- what the target architecture is for balancing deterministic code-owned behavior with AI-driven decisions that currently rely on prose or carry-forward state
In other words: where do you think this kind of operational state should live long term — typed control/config models, task-owned prompts, session/event logs, workspace-local markdown, or something else?
Many thanks in advance for any hints or answers 🙏
Best regards from Bangkok!
Beta Was this translation helpful? Give feedback.
All reactions
Replies: 2 comments
-
Hi Dear Dev. Sorry for missing the discussion.
First of all, the Brain / frontal-lobe component was originally designed to maintain consistency in the Agent's work across multiple conversations. However, in actual usage, we ran into the following issues:
The AI tends to use the frontal lobe to store memory, but in practice this only leads to hallucinations.
In most cases, the frontal-lobe component failed to actually participate in the work.
After moving to the Workspace / Harness architecture, the Agent Runtime mostly comes with its own memory system, which makes the frontal lobe no longer necessary.
For maintaining transaction consistency, the AI tends to rely on the commit log of the Trading-as-Git system, which also aligns with the Coding-to-Trading paradigm shift.
Taking all of the above into account, we have removed the frontal-lobe component, allowing OpenAlice to focus more on running the Trading Harness.
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1
-
Additionally, if you'd like to extend the AI's memory system, we will support injecting new MCPs from external sources when creating a Workspace in the future, and referencing specific memories via MCP. This way, you can losslessly preserve the prefrontal-lobe extensions you've already built, and continue to enjoy the advantages of the Workspace architecture.
Beta Was this translation helpful? Give feedback.
All reactions
-
👍 1