Platform

The platform your agents run on.

The agent platform

Build the agents your enterprise runs on.

Mantle isn’t only the data layer. Design, run, and govern AI agents on top of it, grounded in your knowledge graph and safe by default. This is the part that makes it infrastructure.

agent.run running
Renewal Prep Agent
build_context→ 12 sources, 4.2k tokens, cited
plan→ 3 steps
act→ draft renewal email (Gmail)review
cap 5 credits · kill-switch on · every write reviewed

Describe it, don't build it

Tell the Architect the outcome you want in plain language. It drafts the agent, you confirm, and it ships. No pipelines, no glue code.

Grounded in your business

Every agent runs on your knowledge graph. It reasons on real, cited context from across your systems, not the public internet and not guesses.

Takes action across your apps

Agents don't just read. With governed app-actions they draft and execute in the tools you already use, and every write is reviewed before it lands.

Learns how your work gets done

Agents capture the procedures your team follows and reuse them. Knowledge compounds as skills, so agents get better the more they run.

Governed by default

Hard cost ceilings, a kill-switch, human-in-the-loop review, a contradiction queue, and full run history. Nothing runs unwatched or unbounded.

Runs on schedule, as a team

Trigger agents on a schedule or an event, and let multiple agents coordinate toward a graded outcome, iterating until the work meets the bar.

One-call context assembly

One call assembles the whole context.

Your agent sends a task and a token budget. Mantle returns a ranked, deduplicated, cited package that fits, replacing the gather-and-sift loop agents run today. Fewer calls, fewer tokens, higher relevance.

The usual way4 to 6 calls
  1. 1search_docs(query)
  2. 2fetch(doc_ids)
  3. 3search_crm(account)
  4. 4fetch(records)
  5. 5rerank(candidates)
  6. 6dedupe + trim to budget

Every agent reimplements this loop, burning calls and tokens before it even starts reasoning.

With mantleMCP
build_context({
task: "Prep the Acme renewal call",
token_budget: 4000
})
ranked, deduplicated, cited
context package, within budget
one request · provenance on every fact

One call, not six

Mantle runs the gather, rank, and dedupe server-side, so your agent makes a single request instead of a multi-step loop.

Fits your token budget

You pass a budget; the package comes back within it. No overflow, no padding the prompt with context the model never needed.

Cited, not guessed

Every item carries its source and a quality score, so the agent can ground its answer and you can verify where it came from.

Built for the enterprise

One context layer for every agent in your org.

Connect Drive, Gmail, Slack, Notion, and the rest of your stack once, with Salesforce, Snowflake, and more landing soon. Every internal agent, sales copilot, support assistant, ops automation, data lookup, shares the same context. Permissions and audit logs come standard.

Talk to us about enterprise

MCP-Native

Plug in any MCP-compatible client.

mantle speaks the Model Context Protocol, the open standard from Anthropic. Connect any client. No adapters, no glue code, no vendor lock-in.

Claude
GChatGPT
Gemini
Cursor
VVS Code
Claude Code
Windsurf
LangChain
CrewAI
AAutoGen
+Custom agents

Data Sources

Reads your data where it lives.

Zero-copy connectors that read in place. No ETL, no duplication, no copies.

Available today

Google Drive
Gmail
Slack
Dropbox
Notion

Available on request

PostgreSQL
Google Cloud Storage
S3Amazon S3

Coming soon

SSalesforce
Jira
HubSpot
GitHub
BigQuery
Confluence
Snowflake

For developers

Five minutes. One call. Any stack.

Drop your MCP endpoint into Claude, Cursor, ChatGPT, or any agent framework, and you’re live. No SDK, no migration, no glue code, no ETL pipelines to babysit.

  • From signup to first answer in under five minutes.
  • No SDK, no migration, no pipelines to maintain.
View Documentation
// Type-safe client against the live OpenAPI spec, no SDK required.
// Generate types: npx openapi-typescript https://api.mantleai.dev/openapi.json -o mantle.d.ts
import createClient from "openapi-fetch"
import type { paths } from "./mantle.d.ts"
const client = createClient<paths>({
baseUrl: "https://api.mantleai.dev",
headers: { Authorization: `Bearer ${process.env.MANTLE_API_KEY}` },
})
const { data } = await client.POST("/v1/search", {
body: { query: "Full relationship with Acme Corp", limit: 10 },
})
console.log(data?.[0]?.score) // → e.g. 0.91

Works with everything

TypeScriptPythonREST APIClaudeClaude CodeCursorChatGPTVercel AI SDKLangChainLangGraphCrewAIOpenAI SDKMastraZapiern8nPipecat

Imagine every agent in your company knowing what your best employee knows. That is what Mantle is for.

Get started

Your AI deserves
deeper context.

Connect your first source in under five minutes, or talk to us about an enterprise rollout.

By signing up, you agree to our Terms and Privacy Policy.