setup-burden.md

Setup Burden

Low Burden

Solution	What you do	What you do not operate
Membase	Create an account, use dashboard chat, optionally connect an AI tool/plugin, import chat history, Gmail/Google Calendar/Slack, Notion, or Wiki material.	Storage, indexing, hosted retrieval, background organization, MCP server hosting, and connector plumbing; Notion sync, Drive, and GitHub are coming soon unless verified.
OpenHuman	Install desktop app, onboard, connect selected apps, choose managed or custom providers.	Local memory plumbing and default integration orchestration.
Hjarni	Add Hjarni as a custom connector / remote MCP server in Claude or ChatGPT via OAuth (official: under 2 minutes), then write notes or have your agent write them.	Storage, indexing, search, MCP server hosting, and the REST API. Free plan is capped at 25 notes, containers, and tags; a paid plan removes the cap.
ChatGPT Memory	Turn memory settings on or off and manage saved memories.	External connectors, MCP, vector DBs, or custom collectors.
NotebookLM	Create a notebook and add sources.	Indexing infrastructure or custom retrieval code.

Medium Burden

Solution	What you operate	Main risk
Supermemory	MCP/API auth, optional project scoping, connectors, API usage.	Connector status, project boundaries, and source deletion semantics.
Hyperspell	App/API keys, user tokens, Hyperspell Connect, source selection, metadata/collection design, MCP or SDK integration.	Private beta availability and app-owned governance/review UI.
Honcho	Hosted MCP/API key or self-hosted FastAPI server, SDK/MCP integration, peer/session design, provider keys for self-hosting.	Agent memory quality depends on integration design; self-hosting adds backend operations.
Mnemosyne	Python package, local SQLite data path, MCP/SDK/Hermes plugin configuration, embedding settings, memory banks, and consolidation cadence.	Easy to start, but retrieval quality and write-back safety depend on agent integration and memory-scope design.
taOSmd	Source install of the Python package, a local embedding model, a local LLM (Ollama or RKLLM), the data directory, and optional MCP or HTTP serve configuration.	Easy to start from source, but there is no PyPI package yet, and retrieval and extraction quality depend on the local models and configuration.
Vestige	One npm install of the binary, an MCP config entry, a one-time embedding-model download, and the agent-memory protocol wiring; Intel Mac needs a Homebrew ONNX Runtime path.	Easy to start, but retrieval, consolidation, and suppression quality depend on the local model, memory-scope design, and how the agent protocol is wired.
Mem0/OpenMemory	API keys, SDK integration, hosted or self-hosted stack.	Memory scope design, governance, and retrieval tuning.
Zep/Graphiti	App integration, user/session/group model, graph ingestion, graph backend, LLM/embedding provider.	Requires product engineering rather than end-user setup.
Cognee	Python package/SDK setup, optional MCP client config, optional Docker or API/Cloud mode, graph processing.	Separate standalone instances vs shared API mode can fragment memory.
Hermes Agent + LLM Wiki	Hermes install/config, WIKI_PATH, source curation, Markdown review, lint/maintenance cadence.	Easy to start, but quality depends on agent discipline and user review.
obsidian-wiki	Python package/setup, vault path, agent/model selection, source review, backups/sync, and a maintenance cadence.	Easy to install, but knowledge quality and safe write-back depend on the executing agent and user review.
Hermes Agent + Obsidian + Honcho	PostgreSQL + Redis setup, Honcho init, Hermes config, Obsidian vault creation, AgentMail API key, skill wiring.	Higher setup burden (~60 min); multiple components to integrate and maintain. Core components are free/OSS; AgentMail is a hosted service.
Khoj	Cloud or self-host install, source configuration, indexing.	Self-hosting and source freshness need active management.

High Burden

Solution	What you operate	Main risk
GBrain	Local CLI/init, brain repo, import/sync/embed, dream/autopilot, stdio/HTTP MCP, recipes or collectors; company brain adds source/OAuth/database design.	Official personal setup targets about 30 minutes, but broad active-context coverage is still operated by the user.
Obsidian/Logseq + AI bridge	Vault hygiene, plugins or bridges, local model/API choices, sync and backups.	Strong ownership, but AI behavior depends on the bridge you build.

Practical Rule

Start with the lowest setup burden that satisfies your privacy and portability needs. Move to a local/self-hosted stack only when you have a real reason to own storage, indexing, or graph construction.