Dakera vs Supermemory: connectors to your apps, or a memory engine you own
Supermemory is a context platform: it connects to the apps where your knowledge already lives and keeps user profiles current. Dakera is a memory engine that runs on your hardware, where the data you bring is processed on-device. This page compares how data gets in and where it is processed.
Reviewed October 2026
Bring your data: two different ways in
Supermemory's strength is reach: connectors pull content from the tools your team already uses. Dakera's strength is processing: whatever you send is embedded, transcribed, indexed and ranked on the machine you run it on.
| Capability | Supermemory | Dakera v0.12.0 |
|---|---|---|
| Connectors to your appsGoogle Drive, Gmail, Notion, OneDrive, GitHub, Granola and a web crawler, with real-time webhooks plus a 4-hour sync | Yes | No |
| Documents and mediaSupermemory extracts text, PDFs, Office files, images and video (up to 50 MB per file). Dakera stores attachments (25 MiB by default) next to the memory that cites them | Yes | Yes |
| No data leaves your networkEmbeddings, reranking and entity extraction run in the process; the image ships its models and starts with no download | Partly | Yes |
Supermemory's local server and Enterprise plan also run on your infrastructure; the local server uses a model you configure (hosted providers or Ollama) and is licensed for up to 10,000 documents.
The same call, two shapes
Both expose a small add-and-search surface. Supermemory scopes by container_tag; Dakera scopes by agent, session and namespace, with API keys that can be pinned to a namespace.
from supermemory import Supermemory
client = Supermemory()
client.add(
content="Prefers dark mode",
container_tag="user_123",
)
r = client.search.memories(
q="UI preferences",
container_tag="user_123",
)from dakera import DakeraClient
client = DakeraClient(
base_url=URL, api_key=KEY)
client.store_memory(
agent_id="user_123",
content="Prefers dark mode",
)
r = client.recall(
agent_id="user_123",
query="UI preferences",
)Python SDKs: supermemory and dakera. Dakera also has TypeScript, Go and Rust SDKs.
Walkthrough: a recorded stand-up becomes a memory
A v0.12.0 server with attachments enabled turns audio into searchable memory without a transcription API in the loop.
# 1. Upload a recording (stays on your server)
curl -X POST $URL/v1/namespaces/uploads/attachments \
-H "Authorization: Bearer $KEY" \
-H "Content-Type: audio/wav" --data-binary @standup.wav
# 2. Transcribe on-device: the transcript becomes a memory
curl -X POST "$URL/v1/namespaces/uploads/attachments/$REF/transcribe" \
-H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
-d '{"agent_id": "team-bot", "tags": ["standup"]}'
# 3. Recall it by meaning, in any supported language
# client.recall(agent_id="team-bot", query="what did we decide?")The recording, the transcript, the embedding and the index all stay on the server. Attachments and speech to text are opt-in switches described in Multimodal memory; search languages are in Multilingual search.
Side by side
| Supermemory | Dakera | |
|---|---|---|
| Positioning | Memory and context engine: memory, RAG, user profiles, connectors and file processing in one API | Self-hosted memory engine for agents |
| Memory model | Graph-based memory that tracks how facts change and expire; static and dynamic user profiles | Memories with an on-device knowledge graph (5 edge types), 6 decay curves and spaced-repetition TTL, plus consolidation and deduplication |
| Retrieval | Hybrid search over memory and documents; optional reranking and query rewriting | HNSW vector search plus BM25 with RRF fusion, then a cross-encoder rerank; per-request rerank_candidates and a rerank_report on every response |
| Hosting | Managed cloud; local single-binary server; Enterprise with air-gapped self-hosting | Self-hosted, with multi-node high availability and replicated writes |
| Pricing | Free ($5 credits), Pro $19/mo, Max $100/mo, Scale $399/mo, Enterprise custom | No usage fees; you pay for your own server |
| Agent tooling | Hosted MCP server and plugins for Claude Code, Cursor, Codex and OpenCode | MCP server with 14 core tools (90+ via profiles); Dashboard with a Recall lab that shows why each memory ranked where it did |
| Licence | MIT repository; Enterprise is proprietary | SDKs, CLI and MCP server are MIT; the server binary is proprietary |
| LoCoMo | Publishes its own rankings and LongMemEval results; see its research page | 88.2% Recall@20: does retrieval surface the answer (1,536 questions). Published QA-accuracy figures measure whether an LLM answers correctly, so the two are different measurements. Method |
Operating it yourself
If the reason to leave a platform is control, v0.12.0 adds the operator tools: /health/live and /health/ready probes, dakera --check-config, dakera models list|pull|prune, encrypted and compressed backups, AES-256-GCM at rest once a key is configured, and dakera downgrade to convert data back to v0.11.108. The Docker image (amd64, CPU) is 783 MB with default models included and was ready 4.4 s after start. Keeping data on infrastructure you control also simplifies data-residency and compliance reviews.
When to choose each
Choose Supermemory if
- Your knowledge lives in Drive, Gmail, Notion, OneDrive or GitHub and you want it synced in without writing ingestion code.
- Always-current user profiles are central to the product.
- You want a managed platform with tiered pricing and plugins for the major coding agents.
- You like an MIT-licensed repository you can read and a free local server to prototype on.
Choose Dakera if
- Memory content, audio and page images must be processed on hardware you control, with no external model API.
- Your users write in many languages, or your data includes recordings and scanned pages.
- You want flat, predictable cost instead of metered tokens.
- You want one engine for decay, graph, sessions and ranking, shared by every agent through REST, four SDKs or MCP.
Prefer managed? Dakera Cloud is coming. Join the waitlist.
Frequently asked questions
Does Dakera have connectors to Google Drive, Gmail or Notion?
No. Dakera does not sync from SaaS apps; your code, agent or integration sends content through the REST API, SDKs or MCP. If built-in connectors are the priority, Supermemory is the better fit.
Does Dakera process files and audio?
Yes, as opt-in v0.12.0 features: attachments, on-device speech to text from WAV audio, and visual recall over PNG document pages. Everything runs on your server.
Can I self-host Supermemory?
Yes. It offers a free local single-binary server for one machine, licensed for up to 10,000 documents, and an Enterprise plan with distributed and air-gapped options. The local server uses a model you configure, such as a hosted provider or Ollama.
Does Dakera need an LLM?
No. Embeddings, reranking and entity extraction run on-device with ONNX models shipped in the image, so there is no LLM or external API in the memory path.
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.
Prefer managed hosting? Join the Dakera Cloud waitlist →