Personal Agents

A personal agent is an AI assistant that lives outside the getbased browser tab — on your laptop, in your messenger, or on a self-hosted server — and answers questions about your labs using the same context the in-app chat sees. "What's my vitamin D trend?" in Telegram, "summarize my last labs" in Claude Code, or a scheduled "alert me if my hs-CRP goes above 1.0" on Hermes.

Everything runs over Agent Access — a read-only token gates what agents can see. No raw data leaves the browser, no sync mnemonic is shared.

Supported agents

Any client that speaks the Model Context Protocol works. Tested & documented with:

• Hermes Agent — self-hosted messenger bot (Matrix, Telegram, Signal, Discord, WhatsApp). Persistent, schedule-aware.
• OpenClaw — self-hosted AI assistant you wire into any messenger.
• Claude Code — CLI agent for terminal workflows
• Claude Desktop — native MacOS/Windows app
• Cursor, Cline, Windsurf — IDE-integrated agents

Any other MCP client works too — the adapter is client-agnostic.

How It Works

getbased (browser) ──► Context Gateway ──► your agent (via MCP)
                       (text summary)       (read-only)

1. You toggle Agent Access in getbased Settings → Data. A read-only token is minted.
2. getbased pushes a text summary of your labs, context cards, supplements, and goals to the context gateway on every save.
3. Your agent (Hermes, OpenClaw, Claude Code, …) uses the getbased-agent-stack adapter with your token to read that summary and answer questions.

Setup

1. Enable in getbased

Go to Settings → Data → Agent Access and toggle it on. A read-only token is generated — copy it.

2. Install the agent stack

Linux, one command:

curl -sSL https://getbased.health/install.sh | bash

Installs the MCP adapter, the local RAG knowledge server, and the browser setup dashboard; starts the latter two as systemd user services. Read the script first if you prefer to audit before running — the Interpretive Lens guide lists the sha256sum -c verification command.

macOS / Windows / WSL1, or prefer a manual path:

pipx install --include-deps "getbased-agent-stack[full]"
# or with uv:
# uv tool install --with-executables-from getbased-rag \
#                 --with-executables-from getbased-dashboard \
#                 --with-executables-from getbased-mcp \
#                 "getbased-agent-stack[full]"
getbased-stack init --yes      # writes config, generates API key

If you only want the MCP adapter and nothing else (no RAG), pipx install getbased-mcp is a smaller ~10 MB alternative.

3. Generate client config

Launch the dashboard (or open it via the one-click login URL the installer prints at the end) and copy a paste-ready config block for your client:

# Already running via systemd after install.sh — visit http://127.0.0.1:8323
# Manual install? run:
getbased-dashboard serve       # http://127.0.0.1:8323

Open it in a browser, paste your bearer key at the auth gate, switch to the MCP tab, pick your client from the dropdown (Claude Desktop / Claude Code / Cursor / Cline / Hermes / OpenClaw), click copy. Paste into the client's config file (path shown in the dashboard).

Click Run test to verify the adapter spawns correctly before switching to your agent.

4. Chat

In your agent, ask anything about your labs. The adapter exposes these tools:

Tool	Purpose
getbased_lab_context	Full lab summary — values, ranges, trends, context cards, supplements, goals
getbased_section	One specific section (hormones, biometrics, etc.) or the section index
getbased_list_profiles	List all profiles you track
knowledge_search	Semantic search over your Knowledge Base — requires the local RAG server running (systemd user service on Linux after install.sh, or lens serve manually on other platforms). Degrades gracefully when unavailable.
knowledge_list_libraries	List RAG libraries + which is active
knowledge_activate_library	Switch the active library for subsequent searches
knowledge_stats	Per-source chunk counts for diagnosing missing results
getbased_lens_config	Show the Custom Knowledge Source endpoint + key so the agent stays in sync with what the PWA is configured to query

Running a local knowledge base

getbased-agent-stack[full] includes lens serve — a local RAG server. install.sh already starts it as a systemd user service on Linux; on other platforms, run it manually:

lens serve                   # starts on 127.0.0.1:8322
# ingest via the dashboard's Knowledge tab, or CLI:
lens ingest ~/Documents/research

With lens serve running, knowledge_search is grounded in your own document corpus (research papers, clinical guides, personal notes).

See Interpretive Lens for the full RAG setup — including per-library embedding models (MiniLM for speed, BGE-M3 for quality) and connecting the same server to the in-app AI chat via the External server backend.

Hermes Agent quick-config

# ~/.hermes/config.yaml
mcp_servers:
  getbased:
    command: getbased-mcp
    args: []
    env:
      GETBASED_GATEWAY: https://sync.getbased.health
      GETBASED_TOKEN: your-token-here

Restart: hermes gateway restart. The dashboard's MCP tab generates this exact YAML (with the fields pre-filled) — copy from there to avoid typos.

Security

• The token grants read-only access to a pre-built text summary — no writes, no mutations, no raw data
• Your sync mnemonic is never shared — Agent Access is a separate system
• Revoking (toggle off or regenerate) immediately invalidates the token
• Treat the token like a password — don't commit it to a public repo

Self-hosting the gateway

The context gateway is a lightweight Node.js server. If you self-host your sync relay, the gateway runs alongside it on the same VPS. All data stays on your own infrastructure.