COMPARE · SELF-HOSTING

Dakera vs Mem0 self-hosting: count the moving parts

Q: Does self-hosted Dakera need any API keys?

Only the key that protects your own server ( DAKERA_ROOT_API_KEY ). No LLM or embedding provider key is needed; embeddings, reranking and entity extraction run on-device with the models shipped in the image.

Both projects can run on your own machines. The difference is how many things you stand up, keep patched and connect before the first memory is stored. Here is the inventory for each, taken from their own docs.

Reviewed October 2026

What you run

Mem0's self-hosted server runs as a Docker Compose stack, and its memory extraction and embeddings call model providers. Dakera is a single container that carries its models.

Mem0 self-hosted server3 containers + 2 providers

API servercontainer, port 8888

Postgres + pgvectorcontainer

Dashboardcontainer, port 3000

LLM providermemory extraction; OpenAI by default

Embedding providerOpenAI by default

Dakera1 container

DakeraAPI, storage, embeddings, reranker and entity extraction in one process; models baked into the image

Your infrastructureThird-party API

Mem0 components from its docker-compose file and open-source docs. The LLM and embedder are configurable: Mem0 bundles OpenAI, Anthropic and Gemini providers and the defaults are OpenAI models.

First hour, step by step

The documented path to a running server in each project.

Mem0 self-hostedDakera

Mem0 self-hosted

# From the Mem0 repo, server/ directory
git clone https://github.com/mem0ai/mem0
cd mem0/server

# .env needs POSTGRES_PASSWORD,
# JWT_SECRET and an LLM provider key
make bootstrap
# or: docker compose up -d

# API on :8888, dashboard on :3000

Dakera

docker run -d --name dakera \
  --restart unless-stopped \
  -p 3000:3000 \
  -e DAKERA_ROOT_API_KEY=my-dev-key \
  -e DAKERA_STORAGE=filesystem \
  -v dakera-data:/data \
  -v dakera-models:/app/models \
  ghcr.io/dakera-ai/dakera:0.12.2

curl http://localhost:3000/health/ready

Dakera image and flags are from the Quickstart. /health/ready returns 200 once the models are loaded.

On day one	Mem0 self-hosted	Dakera
Containers	API server, Postgres with pgvector, dashboard	One
Secrets to set	`POSTGRES_PASSWORD`, `JWT_SECRET`, a provider API key	`DAKERA_ROOT_API_KEY`
Model providers	An LLM and an embedder, OpenAI by default	None; models run in the process
Database to operate	Postgres with pgvector	None; data lives on a volume
Platform-only features	Graph memory, decay and temporal reasoning are Platform features, not in open source	Entity graph and 6 decay curves ship in the server

What Dakera v0.12.0 adds for operators

783 MB

CPU image (amd64) with the default models baked in

4.4 s

ready after start, with no download

~101 MB

release binary (x64)

network calls needed to start, so air-gapped hosts work

Know before you start: /health/live and /health/ready probes, and dakera --check-config to validate configuration without starting the server.
Manage models: dakera models list|pull|prune; for other models in an air-gapped setup, seed a models directory from a machine with access. See Models and Docker.
Upgrade and back out: upgrades from v0.11.108 run in place, and dakera downgrade converts data back to what v0.11.108 reads. See Rolling back.
Protect and watch: encrypted and compressed backups, AES-256-GCM at rest once a key is configured, and Prometheus alert rules with a Grafana dashboard in dakera-deploy.
High availability: multi-node HA with replicated writes, for resilience rather than extra capacity.

Multi-arch images cover amd64 and arm64, so the same single container runs on a small home server or a NAS.

When to choose each

Choose Mem0 self-hosted if

You already operate Postgres and a model provider and one more compose stack is routine.
You want to use the same open-source core that Mem0 Platform runs, under Apache 2.0.
You are happy with OpenAI-class providers doing extraction and embedding.
You want a path to Mem0 Platform later.

Choose Dakera if

You want one thing to deploy, monitor and patch.
The host has no outbound network, or must not send memory text to a provider.
You need multilingual or multimodal memory on your own hardware.
You want rollback and config checks as built-in commands.

Prefer managed? Dakera Cloud is coming. Join the waitlist.

Frequently asked questions

Does self-hosted Dakera need any API keys?

Only the key that protects your own server (DAKERA_ROOT_API_KEY). No LLM or embedding provider key is needed; embeddings, reranking and entity extraction run on-device with the models shipped in the image.

Can Dakera run with no network access?

Yes. The default image ships its models, so it starts with no download; it was ready 4.4 s after start in v0.12.0 measurements. Additional models can be seeded from a machine with access.

Does Mem0 self-hosted need an LLM key?

Per the Mem0 setup docs, yes: an LLM provider key is required, with OpenAI as the default. Other providers and local models are configurable.

Is there a production setup for Dakera?

Yes: Docker Compose, Kubernetes and Helm, plus a multi-node high-availability cluster with replicated writes. Profiles live in dakera-deploy; see Deployment.

Self-hosted is free

Run Dakera on your own infrastructure.

One binary with its models baked in and no per-call fees. Start it with one Docker command and see it for yourself.

Run locally in 5 minutes → Read the docs

Prefer managed hosting? Join the Dakera Cloud waitlist →