Hermes + Alani Insights
A GUIDED TOUR OF THE AGENT STACK

Hermes does the work.
Alani Insights keeps the record.

Memory lives in your agent and fades. Alani Insights is the storage and the context underneath: a shared database your agents read, write, and actually understand, and your team works in like a normal tool. We'll walk you through it in eight short stops.

FIRST STOP
01

ZOOM OUT

Agents arrived. The data layer didn't.

In under a year, open-source agents went from curiosities to some of the most active projects on GitHub, running real operations from Mac Minis and cheap cloud servers. The tooling is real. What analysts, researchers, and the model builders themselves keep flagging is everything around the tooling: the state, the storage, the truth.

ANALYST FORECAST

40%+

GARTNER, JUNE 2025

Gartner predicted that over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, or inadequate risk controls.

Read the press release →
PEER-REVIEWED RESEARCH

“A shared cognitive infrastructure”

PAKDD 2026, SPRINGER

How peer-reviewed research describes memory in multi-agent systems: the infrastructure enabling collective intelligence and long-term coordination, with synchronization and access control named as core challenges.

Read the paper →
MODEL BUILDER GUIDANCE

Context rot

ANTHROPIC ENGINEERING

Anthropic documents that model recall degrades as context windows grow, and recommends persistent, structured memory outside the context window for agents that work over long horizons.

Read the guidance →

Three different researchers, three different angles. The same open question runs through all of them: where does an agent's work live?

THE TAKEAWAY Agents are here to stay. The open question is where their work gets stored.

02

THE PROBLEM

Getting it done is not the same as keeping it.

Hermes gives you an agent that works around the clock. Role agents give you a workforce. But everyone who runs agents seriously hits the same wall: if an agent's work isn't stored somewhere organized, it doesn't add up to anything.

Threads drift

Context lives in prompts and chat history. Every long thread degrades, and every new thread starts from zero.

Work doesn't add up

Agents produce prose. Prose piles up in markdown files and message logs where no other agent, and no human, can reliably build on it.

No shared truth

Each agent keeps its own version of reality. Handoffs get lossy, work gets repeated, and the business state lives nowhere.

THE TAKEAWAY More agents means more activity. A shared, organized record is what turns activity into progress.

03

THE ROAD SO FAR

You've probably already tried these.

If you run Hermes, this probably isn't your first attempt at giving your agents somewhere to keep what they know. Most operators walk the same road. Every tool below is good at its actual job, and every one leaves the same gap. Step through your own journey.

WHY IT MADE SENSE

More hands, more work getting done. If one agent is useful, five should be five times as useful.

WHERE IT BROKE

Nothing held the shared state. The agents repeated work, contradicted each other, and no single place knew what was true.

Labor was never the bottleneck. Storage was.

THE TAKEAWAY Six stops, six partial fixes. None of them sits between your agents and your data.

04

THE TWO HALVES

What each one actually does

Every stop on that road solved a piece. Here is the missing one. But first, three words that get mixed up in every conversation about agents:

Memory

What an agent holds in its head. Powerful, personal, and it fades with every new thread.

Context

What an agent is handed for the task in front of it. Only as good as where it comes from.

Storage

Where the truth lives. Organized, permanent, and shared by every agent and every person.

Hermes carries the memory. Alani Insights carries both the storage and the context. Underneath, it is real tables. On top of every field sits a layer of meaning: its description, its logic, how it relates to everything else. Agents don't just read the values, they read what the values mean. That is the context layer.

THE AGENT

Hermes

  • Open-source personal AI agent, built by Nous Research
  • Lives in your messaging apps, with Telegram first in line
  • Splits big jobs across subagents that work in parallel
  • Learns skills as it goes and connects to outside tools over MCP

Great at doing work. It remembers you and its skills, not your business.

STORAGE + CONTEXT

Alani Insights

  • An AI-native database: real tables, typed fields, relationships
  • A context layer on top: every field carries its own description and logic
  • A no-code workspace for humans, MCP read and write for agents
  • Table-scoped API keys: each connection touches only what you allow

Built to store what your business knows, and explain it to your agents.

THE TAKEAWAY Hermes brings the memory. Alani Insights brings the storage and the context. That is the whole stack.

05

THE MAP

Where your agent's data lives

Every Hermes setup eventually runs into the same four questions. Where does context come from? Where do results go? Who is allowed to write what? And what will still be on the record next month? Pick any node to see how this architecture answers them, or trace a real request through the whole stack.

Agents Models Channels & intake Context layer

PEOPLE

delegates and reviews

AGENTS

agents call models to do the thinking

MODEL LABOR

reads and writes over MCP
scoped intake, one table each

INTAKE EDGES & RUNTIME

THE TAKEAWAY You can swap any white box. The gold layer holds the records, the permissions, and the history.

06

THE TOUCHPOINTS

The five connection points

Hermes and Alani Insights connect in five ways. Each one turns what your agents do into records you keep.

Your tables become Hermes tools

Hermes ships full MCP client support, and Alani Insights speaks MCP natively. Connect them once and every table you create surfaces as a tool Hermes can read and write: query contacts, update a deal stage, log an activity. And because every field carries its own description and logic, agents understand what they are writing into. No custom code, no middleware.

THE TAKEAWAY Five connection points, one result: nothing your agents do gets lost.

07

THE PARTS LIST

Everything in the stack, in one place.

Every part of a production agent stack, by role. Filter by layer to see how the pieces divide the work, and notice which one you would never swap.

Your agent can change.
The storage underneath shouldn't.

Swap models when prices move. Reorganize your agents when the work changes. If the structured layer is solid, the business keeps its continuity.

Hermes

Open-source personal AI agent. Gathers context, spawns subagents, reports over your messaging apps.

Paperclip

Optional org-chart manager for teams of agents: titles, bosses, budgets, and approvals.

OpenClaw

Another open-source autonomous agent, often run alongside Hermes on home hardware and VPSes.

Claude + Claude Code

Mission-critical reasoning, with heavy lifting through the CLI.

DeepSeek V4 Pro

Cheaper labor for non-critical work. Same tables, a fraction of the cost.

Telegram

The human access layer. Updates go out, approvals come back.

MCP

The open protocol agents use to connect to Alani Insights tables and other tools.

Alani Insights

Storage plus context. Every field explains itself; every agent reads and writes the same tables.

Table-scoped API keys

Isolation without enterprise IAM. A form's key touches one table and nothing else.

Website contact forms

Write straight into your tables. New information files itself.

GitHub

Where the open-source agents live, and a context input Hermes pulls from.

Landing sites

More inputs the agents read for context.

Mac Minis

Home hardware running agent fleets around the clock.

Hetzner-style VPS

Cheap always-on cloud hosts for agents. First-class in the agent community's own docs.

THE TAKEAWAY Fourteen tools, three jobs: act, think, store. Only one layer does the storing.

08 · GET STARTED

Memory fades. Give your agents storage that doesn't.

Connect Hermes to Alani Insights over MCP, and everything your agents do today becomes context they can use tomorrow.

YOUR FIRST FIVE MINUTES

1

Create your database

Sign up and get your own provisioned database. No schema required on day one.

2

Connect Hermes

Add Alani Insights to Hermes as an MCP server. Your tables become tools.

3

Say the word

Ask Hermes to create a table and add a record. From then on, its work lands there.