Keep agent memory readable and retrievable.

zer0dex pairs a markdown memory index that people can inspect with local semantic retrieval for details an agent should look up before a model call.

View source ↗Install from PyPI ↗Read the docs →

Local first resultafter zer0dex check passes

$ zer0dex seed --source MEMORY.md
$ zer0dex serve --background
$ zer0dex query "Where does Atlas deploy?"

§ 01

A local lookup before a model call

For agent and framework developers

Use zer0dex when a single MEMORY.md has become too large to keep in every prompt, but a vector store alone makes the available knowledge hard to inspect. Keep headings, durable summaries, and pointers in markdown; seed the details into the project-local Chroma store.

Your host calls the loopback POST /query endpoint, then decides whether and how returned text enters a prompt. The package supplies that CLI and server; it does not install an automatic agent hook.

§ 02

Get a first local result

Ollama · no hosted memory service

Install zer0dex, run the prerequisite check, seed one markdown file, start the local server, and query it. The default path uses Ollama on your machine and has no cloud API-key step.

STEP 01Install the local prerequisites

$ pip install zer0dex==0.1.1
# In a separate terminal, if Ollama is not already running:
$ ollama serve
# Back in this terminal:
$ ollama pull nomic-embed-text
$ ollama pull mistral:7b
$ zer0dex check

check exits 0 only when Ollama, both models, mem0ai, ChromaDB, and the Python Ollama client are available. Do not start a second ollama serve when the service already owns its local port.

STEP 02Seed a readable index

$ printf '%s\n' '# Memory' '## Project Atlas' '- Deployment target: staging' > MEMORY.md
$ zer0dex init
$ zer0dex seed --source MEMORY.md

The working directory receives .zer0dex.json and .zer0dex/; seeding reports the extracted-memory count.

STEP 03Retrieve through the local server

$ zer0dex serve --background
$ zer0dex query "Where does Project Atlas deploy?"
$ zer0dex status

The query prints scored memories when one clears the configured threshold. status prints the loopback port, memory count, and ok.

Decision boundary. Retrieval is semantic matching, not a truth check. Treat returned source text as data under the same trust rules as any other prompt context.

§ 03

Use the small HTTP surface

127.0.0.1 only

The server exposes GET /health, POST /query, and POST /add. A host can issue one query before dispatching its model call and keep the returned memories in an explicitly untrusted context field.

$ curl -X POST http://127.0.0.1:18420/query \\
  -H 'Content-Type: application/json' \\
  -d '{"text":"preferred reply style","limit":5}'

Successful queries return a memories array whose entries contain text and score. Scores and ordering come from the local backend and vary with the store and query; they are not a stable ranking contract.

§ 04

Developer preview, with local boundaries

0.1.x

zer0dex 0.1.1 is an Alpha developer preview. It binds only to 127.0.0.1 and has no authentication; do not proxy that port without adding an appropriate boundary. It is neither hosted memory infrastructure nor a compliance, access-control, or governance system.

Read installation, configuration, the HTTP contract, troubleshooting, and compatibility guidance.