nodecode-nexus

Nexus Memory: durable facts on a Qdrant collection, guardrails over them. A Nodecode rewrite of Neboy72/nexus-memory by Neboy72.

The model gets five tools - nexus_recall, nexus_remember, nexus_forget, nexus_guardrail_check, nexus_find_entities - over the nexus collection the original keeps, with the same payload fields, so a cell and the original plugin (or any MCP-connected agent) share one memory. What the collection holds about the newest ask rides every request as one nexus block, so recall needs no call; a finished turn is filed in the background; /nexus shows the store or searches it.

Two things are its own: a memory carries an access level (public, trusted, private) and this cell reads only up to the level its section names, with a memory that has no known level counting as private; and a memory stored under category rule is a protection rule, which nexus_guardrail_check matches against the targets of a destructive command before it runs.

Get it

Tell the agent "add the nexus cell", or run nodecode add nc://nexus.

Set it up

A Qdrant you run and an embeddings endpoint of the model the collection was built with:

{
  "nexus": {
    "url": "http://127.0.0.1:6333",
    "collection": "nexus",
    "embedding_url": "http://127.0.0.1:11434/api/embed",
    "embedding_model": "qwen3-embedding:0.6b",
    "api_key_env": "QDRANT_API_KEY",
    "access_level": "private"
  }
}

embedding_url is the one required member: a vector has to come from somewhere. Any OpenAI-compatible /v1/embeddings endpoint answers it as well as Ollama's /api/embed, and its model must be the one the collection's vectors were made with - the original's local model is 1024 dimensions. A Qdrant Cloud store takes its key from api_key (or api_key_env, or QDRANT_API_KEY in the environment). Nothing else is contacted: the memories, the vectors and the ids stay in your store.

Member	Default	What it does
`url`	`http://127.0.0.1:6333`	the Qdrant REST base URL
`api_key` / `api_key_env`	unset	the Qdrant API key, when the store asks for one
`collection`	`nexus`	the collection the memories live in
`vector_name`	empty	the named vector, when the collection has one
`distance`	`Cosine`	what a collection this cell creates uses
`embedding_url`	-	required: the embeddings endpoint
`embedding_model`	`qwen3-embedding:0.6b`	the model that endpoint serves
`access_level`	`private`	the level this agent reads up to and writes at
`agent`	`nodecode`	the name stamped on what this cell writes
`scope`	`default`	the scope a memory is filed under
`recall`	`true`	recall for the newest ask before each turn
`recall_limit`	`5`	the most memories one recall or search answers
`capture`	`true`	file each finished turn in the background
`max_chars`	`400`	how much of a captured turn is stored
`retry_seconds`	`60`	how long a store that just failed is left alone
`timeout`	`20`	one request's ceiling, in seconds

Use it

nexus_remember {text: "Prefers green tea over coffee", category: "preference"}
nexus_recall   {query: "what does the user drink"}
nexus_remember {text: "Never delete /srv/protected - production data", category: "rule"}
nexus_guardrail_check {command: "rm -rf /srv/protected"}   -> verdict block, with the rule
nexus_forget   {memory_id: "..."}

From eval the same verbs are functions: (nexus:recall "what does the user drink"), (nexus:remember "..." :category "rule"), (nexus:status), and (help :nexus) is the primer the model reads.