AI teammates for finance operations.
Reconcile payment sources against the ledger, explain any variances, and prep the reconciliation note.
Build your AI teammate in 4 steps.
See how a finance teammate connects ledger history, transactions, reconciliations, and sign-off without rebuilding context each run.
Connect your systems.
Connectors stream your tools into memory. Models, sandboxes, and devices are independent layers — text, image, speech, and more, credentials gateway-side. Flip the tabs to wire each one.
Connectors ↗Model your business.
Declare the people, business objects, and relationships your agent needs. Every fact from the connectors above converges on the right record, so any teammate reads the same picture.
Memory ↗Define the work.
A trigger — a schedule, or reactive to connector events — plus a plain-language prompt. Once defined it runs unprompted: scans memory, matches the trigger, and surfaces the work with the evidence attached.
Behaviors ↗“Each morning, reconcile every against and . When a explains a variance, attach it and ping the before close.”
Talk to your teammate everywhere.
Delegate in chat — Slack, Teams, WhatsApp, the API, or MCP. The agent proposes with buttons; you review, edit, and approve. Approving triggers a run that hands the work off to the right person.
Platforms ↗Variance flagged · Account 4100
- $12,480 variance vs payouts
- 3 refunds settled after cutoff
- Same 3-day lag as September (STR-44)
Drafted the month-end recon note for Maya. Review before I post it:
A framework for AI teammates you own.
Not hand-rolled scripts. Not a rented employee. An open-source backend — connectors, shared memory, sandboxed execution, and observability in one — for teammates that watch, remember, and act. Yours to self-host and scope per person. See how it compares →
Everything an agent runs on.
Four swappable runtime layers — inference, execution, memory, and connectors — with evals, deployment, and observability built in. Replace any runtime layer without touching the others.







The same agent, in code.
The use case above is one project: connections, entity types, behaviors, and agent configuration. Inspect each piece or let your coding agent generate it.
import {
connectorFromFile,
defineAgent,
defineConfig,
defineEntityType,
defineRelationshipType,
defineBehavior,
reactionFromFile,
secret,
} from "@lobu/cli/config";
import type QuickBooksTransactionsConnector from "./quickbooks-transactions.connector.ts";
import type reconciliationMonitorReaction from "./reconciliation-monitor.reaction.ts";
const finance = defineAgent({
id: "finance",
name: "finance",
description:
"Help finance teams reconcile data, explain variance, and prepare reporting runs",
dir: ".",
providers: [
{
id: "anthropic",
model: "claude/sonnet-4-5",
key: secret("ANTHROPIC_API_KEY"),
},
],
network: {
allowed: [
"github.com",
".github.com",
".githubusercontent.com",
"registry.npmjs.org",
".npmjs.org",
],
},
});
// entity types and relationships defined here…
const reconciliationMonitor = defineBehavior({
agent: finance,
slug: "reconciliation-monitor",
name: "Reconciliation monitor",
triggers: [{ kind: "schedule", cron: "0 6 * * 1-5" }],
notification: { priority: "high", channel: "both" },
tags: ["finance", "reconciliation", "daily"],
minCooldownSeconds: 3600,
reaction: reactionFromFile<typeof reconciliationMonitorReaction>(
"./reconciliation-monitor.reaction.ts"
),
prompt:
"Check accounts for unreconciled transactions, new variances, and approaching reporting deadlines. Lead with exceptions that need review.\n",
});
export default defineConfig({
connectors: [
connectorFromFile<typeof QuickBooksTransactionsConnector>(
"./quickbooks-transactions.connector.ts"
),
],
org: "finance",
orgName: "Finance",
orgDescription:
"Help finance teams reconcile data, explain variance, and prepare reporting runs",
agents: [finance],
entities: [account, report, transaction, variance],
relationships: [createsVariance, reconcilesTo, summarizedIn],
behaviors: [reconciliationMonitor],
});Explore agent workflows.
Each example shows the sources, the memory, and the action for one AI teammate. Same loop (connect, watch, act) pointed at a different job.
Tracks account health, rollout progress, and renewal signals across the customer base.
Reviews incoming contracts, summarizes risk, and surfaces missing protections before sign-off.
Reconciles data across systems, explains variance, and prepares recurring reporting runs.
Declarative. Yours. In your repo.
Every teammate above is a lobu.config.ts — desired state you version, review, and own.lobu validate checks it, lobu applyships it, lobu run boots it. No magic, no lock-in.
Local, self-hosted, or managed.
The same project runs on your laptop, in your cloud, or fully managed by lobu. Your data and controls stay wherever you need them.
Run in your cloud.
Docker, a cloud VM, or Kubernetes when data and controls need to stay with you.
Let lobu run it.
The same project with managed isolation, secrets, and upgrades, nothing to operate.
Questions a technical buyer asks first.
Latest blog posts
The Context Layer Is an Event Store
Tasman says no product can build your context layer because the content doesn't exist. Half right. The governed why is a versioned, access-controlled event store that mirrors the warehouse instead of replacing it — and I built the Kelder scenario to find where that holds and where it doesn't.
Read post →The Agent Loop Is the New SaaS
The business logic SaaS sold you is now an agent loop that watches your data and acts. Here is how to build and own one with Lobu.
Read post →Shopify's Aquifer, in the Open
Shopify treats an agent's corpus as the compounding asset. Lobu keeps the signal—not the chat—and makes that architecture work across companies.
Read post →