"Zustand for state" (0.95)
"no Redux" (0.95)
query: "state management"
Week 4 Continue had zero context from Week 1. The team convention was retrieved from MnemeBrain's persistent belief graph.
Why Continue.dev + MnemeBrain
Open source meets open sourceContinue.dev's MCP support makes this integration clean: register MnemeBrain as an MCP server in ~/.continue/config.json, and Continue gains 13 new memory tools it can call mid-chat. The plugin architecture means adding memory doesn't change how you use Continue — it just makes it better.
Continue works with any LLM — OpenAI, Anthropic, local Ollama models, even your own hosted endpoints. MnemeBrain works alongside all of them: the belief graph is model-agnostic. Switch your LLM tomorrow and your project memory stays intact.
Install the backend
Local, no cloud dependenciespip install mnemebrain-lite mnemebrain-lite serve # → MnemeBrain running at http://localhost:8000
Keep this terminal open while using Continue. For automatic startup, add to your shell profile:
mnemebrain-lite serve &> /tmp/mnemebrain.log &
Clone the MCP bridge
The subprocess Continue spawns on startupgit clone https://github.com/mnemebrain/mnemebrain-mcp cd mnemebrain-mcp uv sync pwd # note this path
pip install uv. It's a fast Python package runner required by the MCP bridge.
Configure Continue.dev
One block in config.jsonOpen ~/.continue/config.json and add the mcpServers block:
{
"models": [ ... ],
"mcpServers": [
{
"name": "mnemebrain",
"command": "uv",
"args": [
"--directory",
"/path/to/mnemebrain-mcp",
"run",
"mnemebrain-mcp"
],
"env": {
"MNEMEBRAIN_URL": "http://localhost:8000"
}
}
]
}
Replace /path/to/mnemebrain-mcp with the path from the pwd command above. Save and reload VS Code.
~/.continue/config.json. The MCP configuration is identical — no separate setup needed for IntelliJ, PyCharm, or WebStorm.
.continue/config.json to your project root (MCP entry only, without personal LLM keys) and commit it. Team members get shared project memory configuration automatically.
Use persistent memory in Continue
The patterns developers use mostStore a code style convention:
Remember: all React components use functional style with hooks. No class components. Prefer composition over inheritance throughout.
Store a dependency decision:
We use React Query for all data fetching. Never fetch directly in components — always go through a custom hook that wraps React Query.
Ask what Continue remembers about the project:
What conventions do we follow for data fetching?
Contradiction detected when patterns change:
Week 6: "we've switched to SWR for the new dashboard"
truth_state: BOTH ← coexisting beliefs, conflict surfaced
belief_1: "React Query" confidence: 0.95 → 0.33
belief_2: "SWR (dashboard)" confidence: 0.89
Continue: ⚡ Data fetching conflict. React Query stored Week 1, SWR now mentioned. Is this a full migration or scoped to the dashboard only?
Most retrieval-based memory systems cannot represent contradictions explicitly — they silently overwrite the old fact. MnemeBrain creates an ATTACKS edge, flags truth_state: BOTH, and lets Continue surface the conflict before giving you advice based on stale data.
What you've built
Continue.dev now has a persistent belief graph that survives across every VS Code session. Code style, library choices, architectural patterns, and team conventions are stored, tracked for contradictions, and retrieved semantically — using whichever LLM you already have configured in Continue.
The belief graph is model-agnostic: switch between GPT-4, Claude, or a local Ollama model without losing any stored beliefs. Typical belief lookup is <5ms on a local Kuzu store with ~100k beliefs.