Install¶

Summary

RelyLoop installs as a Docker Compose stack — clone the repo, run make up, and the whole system (API, worker, Postgres, Redis, Elasticsearch, OpenSearch) comes up on one machine. There is no pip install; RelyLoop is a deployed stack, not a library.

System requirements¶

Requirement	Minimum	Recommended
Docker	24+ with Compose v2	latest stable
RAM	16 GB	32 GB
Free disk	~16 GB	32 GB+
OS	Linux or macOS (x86-64 / arm64)	—

Elasticsearch and OpenSearch each take ~1 GB of heap by default. If you index a large corpus, bump ES_HEAP_SIZE in .env.

Supported engine versions¶

Engine	Versions	Status
Elasticsearch	8.11+ / 9.x	shipped (MVP1)
OpenSearch	2.x / 3.x	shipped (MVP1)
Apache Solr	9.x / 10.x	shipped (MVP2)

All three are reached through a single SearchAdapter Protocol — the same study workflow runs unchanged across engines.

Install¶

git clone https://github.com/SoundMindsAI/relyloop.git
cd relyloop

make up          # generates secrets, builds images, brings up the stack (~90s cold)
make migrate     # apply the Alembic migration chain

make up runs scripts/install.sh, which auto-generates the required local secrets as mounted files (never bare environment variables) and then runs docker compose up -d. When it finishes, the UI is at http://localhost:3000 and the API at http://localhost:8000.

An OpenAI key is optional to boot

RelyLoop starts without an LLM key — /healthz simply reports openai: missing_key and the LLM-dependent features (judgment generation, digests, the chat agent) wait until you configure one. Only the Postgres password is boot-blocking.

Configure an LLM endpoint¶

The entire LLM integration surface is one environment variable, OPENAI_BASE_URL, pointing the openai SDK at any OpenAI-compatible endpoint:

OpenAI cloudOllama (air-gapped)vLLM / LM Studio / TGI

# .env
OPENAI_BASE_URL=https://api.openai.com/v1
# plus an API key mounted as a Docker secret (see the tutorial)

# .env — nothing leaves the host
OPENAI_BASE_URL=http://host.docker.internal:11434/v1
OPENAI_MODEL=llama3.1

# .env — any OpenAI-compatible server
OPENAI_BASE_URL=http://your-host:8000/v1

For Bedrock, Vertex, or Anthropic-native, put a LiteLLM proxy or OpenRouter in front and point OPENAI_BASE_URL at it. Full matrix: LLM endpoint setup.

Next step¶

Head to the Quickstart to seed sample data and open the chat agent, then walk a full study end-to-end in Your First Optimization Loop.