We are building a semantic context layer for AI agents (pre-release)
Hey Indie Hackers,
We're the team building Mantle, a semantic context layer for AI agents.
## The problem we kept running into
For the last several years we've been building data infrastructure, and we've watched every "enterprise AI agent" project fail for the same reason: the agents were smart, but they had no context. Or worse, they had partial context that looked complete enough to be dangerous.
The usual failure mode:
1. You connect the agent to six or seven enterprise systems.
2. The agent pulls data from each of them.
3. It has no idea that "John Smith" in Salesforce is the same person as "jsmith@acme.com" in Slack.
4. It doesn't reconcile the fact that Snowflake says the deal is $42M and the deck says $41.8M.
5. It returns a confident answer based on incomplete data. The user either catches it (best case) or makes a decision based on it (worst case).
The bottleneck was never the LLM. It was that nothing in the stack actually understood what it was pulling.
## What we're building
Mantle is trying to be the reasoning layer between raw sources and the agent. Three things make it different from the "AI data" category that exists today:
1. Zero-copy architecture. Connectors read the source of truth in place. Nothing gets duplicated into a vector DB or warehouse. Your data stays where it lives.
2. Entity resolution as a first-class primitive. The knowledge graph is built from resolved entities, not raw records. "Acme Corp" in six different systems collapses into one node with a confidence score.
3. MCP-native. Responses come back as Model Context Protocol messages with provenance, quality scores, and reconciled values. Any MCP-compatible agent framework can consume them directly.
## Honest status
Pre-release. No paying customers, no SOC 2 cert, connector library is being rolled out one at a time. We'd rather ship a solid Salesforce + Slack + Snowflake loop than claim 40+ connectors that half-work.
If any of this resonates, or if you think we're wrong about entity resolution being the missing piece, we'd genuinely love to hear it in the comments.
Waitlist is at https://mantleai.dev if you want to follow along.
Happy to answer anything.
