by mmTheBest
AI research workspace with a persistent paper database, structured evidence extraction, hybrid search, and proposal workflows.
# Add to your Claude Code skills
git clone https://github.com/mmTheBest/arxieAn AI research workspace backed by a persistent paper database.
Arxie is a self-hostable research system for serious literature work. It combines:
This release branch ships the v0.2.0 product surface: persistent corpora, saved studies, structured evidence, comparison workflows, provider-backed ingest, and a browser study workspace at /app.
No comments yet. Be the first to share your thoughts!
git clone https://github.com/mmTheBest/arxie.git
cd arxie
python -m venv .venv
source .venv/bin/activate
pip install -e .
If you specifically want local embedding-model dependencies on the host, install:
pip install -e .[local-embeddings]
cp .env.example .env
Set at least:
OPENAI_API_KEYFor the full self-hosted stack, .env.example also includes the Paperbase runtime variables for PostgreSQL, Elasticsearch, Redis, MinIO-compatible object storage, queue dispatch, cache lifecycle, and semantic search configuration.
Important runtime defaults:
PAPERBASE_WORKER_QUEUE_BACKEND=redis in the shipped server stackPAPERBASE_OBJECT_STORE_BACKEND=s3 in the shipped server stackPAPERBASE_EMBEDDING_PROVIDER=openai for production semantic retrievalIf you intentionally want a lighter local process mode, you can switch to:
PAPERBASE_WORKER_QUEUE_BACKEND=dbPAPERBASE_OBJECT_STORE_BACKEND=filesystemPAPERBASE_EMBEDDING_PROVIDER=deterministicStart infrastructure:
docker compose -f infra/docker-compose.paperbase.yml up -d postgres elasticsearch minio redis
Apply schema migrations:
docker compose -f infra/docker-compose.paperbase.yml run --rm paperbase-migrate
Start the API and worker:
docker compose -f infra/docker-compose.paperbase.yml up -d paperbase-api paperbase-worker
If you want a more app-like local workflow, use the launcher command:
arxie-local run
That boots the lighter single-user local stack, waits for readiness, and opens
/app. By default it starts PostgreSQL, MinIO, Redis, the API, and the worker.
If you explicitly want the heavier backend-search service too, use:
arxie-local run --with-search
Other useful shortcuts:
arxie-local open
arxie-local down
arxie-local run --rebuild
arxie-local install-shortcut
arxie-local install-shortcut writes a double-clickable Arxie.command launcher
to your Desktop by default.
On the first launch, Arxie may need a few minutes to start Colima, build the application images, and boot the local stack. The shipped local Compose profile is tuned for a single-user machine, including a smaller Elasticsearch heap so the default stack can run on modest laptop memory.
For the single-user local path, Arxie does not hard-block readiness on the
search backend. The default launcher path skips Elasticsearch entirely so parse,
extraction, and the dashboard remain reliable on a modest laptop. The local
Compose services mount the current src/ and services/ directories, so code
updates are picked up on restart. Use arxie-local run --rebuild only after
dependency or Dockerfile changes. If you later start Arxie with --with-search,
the workspace can use the backend search surface when Elasticsearch is healthy.
If you prefer running the services without Compose:
paperbase-db upgrade
paperbase-api
paperbase-worker
In this mode, the default .env.example still points at the self-hosted stack.
If you want a no-MinIO/no-Redis local run, switch the queue, object-store, and
embedding settings as described above before launching the processes.
Useful make targets:
make paperbase-db-upgrade
make paperbase-api
make paperbase-worker
make paperbase-compose-config
src/ra/ Assistant workflows, CLI, and legacy REST API
src/paperbase/ Canonical schema, ingest, parse, extract, search
services/paperbase_api/ Browser-facing corpus API and UI
services/paperbase_worker/ Background job execution
infra/ Self-hosting stack and environment files
Detailed architecture, planning notes, tests, and internal developer process files live on the private development branch, not on this release branch.
MIT