Give AI the context behind your business.
Meerkats is the context layer AI systems run on. It connects your systems, defines every entity and metric once, keeps business state current, holds your rules, and exposes governed actions. Agents read it over MCP and API instead of reasoning over raw data.
HubSpot
Shopify
Google AdsHubSpotDeal 4471 · blocker openShopifyOrder #48211 · $184Google AdsROAS 2.8 · your definitionAI can reach your data. It still doesn’t know your business.
Out of the box an agent doesn’t know what entities exist, how they relate, which metric definition is right, what happened before, what the state is now, which rules apply, or what it may change. It rebuilds all of that from raw rows on every call.
One context layer. Every agent reads it.
Meerkats serves your business context over MCP and API. Claude, Codex, the agents you build, your automations and your internal apps answer from the same definitions, state and rules.
Keep what you’ve built. Agents, n8n workflows and internal apps connect over MCP or the API.
Declare a responsibility once. It runs with state, inside guardrails.
A workflow in Meerkats is a typed contract, not a prompt: the trigger, the condition and its window, the evidence to gather, the risk class, the caps, and who approves. It remembers its last run and stops for a person before anything changes.
Build the systems your organisation needs. One foundation underneath.
A CRM with its own queue and rules. An intelligence layer across platforms with one metric catalog. An internal app for one team. Each with its own permissions, all reading the same context.
Why a context layer makes agents cheap, fast and accurate.
In Meerkats the model does three things: classify the request, fill a typed contract, write the explanation. Everything else is code.
Against an agent reasoning over raw platform data. Context is pre-materialised, only the knowledge a task needs is loaded, and a small fast model is enough.
- Prepared context, not raw payloads
- Tool lists stay out of the prompt
- Only what the task needs is loaded
- A small model is enough
Names are resolved by the catalog in code. Numbers are computed in code. There is no multi-turn tool exploration, so a monitor on a tight schedule finishes in time.
- One small router call
- Deterministic name resolution
- Queries run where the data lives
- No agent loops
Metrics are defined once. Unknown names are refused, never guessed. Every value shows where it came from and how fresh it is. Actions execute exactly as approved, or not at all.
- Typed, schema-validated tasks
- Catalog refuses unknown names
- Source and freshness on every value
- Read-back after every action
Start with the questions your teams already ask.
Every system starts as a question someone asks each week. Pick one and watch the answer assemble from your context.
Your AI remembers what happened.
Every change and every result is kept in order, with its source — so when a number moves at 3:40 PM, the explanation can start at 9:00 AM.
Conversion rate fell at 3:40 PM after two earlier changes — the target was raised at 9:00 AM and the checkout was updated at 1:20 PM.
Your business changes. Every system stays in sync.
Change a target, a rule or a cap in the workspace and every agent reads it on its next run — no prompts rewritten. Every value carries how fresh it is, and every action re-reads live state first.
Know what your AI knew, decided and did.
Every run keeps its inputs with their sync time, the rule that fired, the plan, who approved it, the exact change and the read-back. Open any step.
You decide how much responsibility AI gets.
Actions come from a fixed, permissioned list. Risk is computed server-side, caps apply even to approved actions, and high-risk changes ask twice.
Six layers. One shared context.
Data flows in, the semantic and context layers give it meaning and state, workflows act on it, agents read it, and governance records every decision.
Ad platforms, marketplaces, store, payments, CRM, team channels, documents, warehouses. Synced on a schedule, re-read live before any action.
One catalog maps platform labels to your definitions. Unknown names are refused, never guessed.
What exists, how it connects, what happened, what is true now, and what is allowed. Loaded by agents instead of rebuilt on every call.
Typed contracts with triggers, windows, caps and approvals. They carry state across runs and stop for a person before anything changes.
Claude, Codex, your agents and apps over MCP and API. Actions bind to a registry of typed, permissioned operations.
Who can do what, what was approved, and a record of every run with its inputs and read-back.
Connect the systems your business already runs on.
Advertising, marketplaces, store, payments, CRM, team channels, documents, and any warehouse or API. Reads and writes, within the permissions you grant. Hover to see what each one adds.
MetaCampaign performanceGoogle AdsSearch & campaign performance
Amazon AdsSponsored ads performance
Flipkart AdsMarketplace ad performance
TikTok AdsVideo campaign performance
Walmart ConnectRetail media performance
BlinkitQuick-commerce ads & availability
ZeptoQuick-commerce ads & availability
InstamartQuick-commerce ads & availabilityShopifyOrders & revenue
RazorpayPayments & refunds
GA4Site behaviour & conversionsHubSpotLeads, deals & pipelineInventoryStock levelsWarehousePostgres, BigQuery, Snowflake tablesRoles, permissions and isolated workspaces.
Each workspace isolates its connections, context, rules, people and history. API keys are scoped per app and per workspace, read or write. Agents act with the approving user’s rights, never more.
AI agents can access your data. Meerkats gives them an understanding of your business.
One context layer every agent, workflow and app reads.
Give AI the context behind your business.
HubSpotDeal 4471 · stage 3Questions, answered.
Still wondering whether Meerkats fits your stack? Start with one workflow and see it run on your own business context.
Meerkats is the context layer AI systems run on. It connects the systems where your business data and knowledge live, defines every entity, metric and relationship once, keeps current state and history, holds your rules, and exposes a fixed set of governed actions. Agents read that context over MCP and API instead of reasoning over raw data.