AI infrastructure company

Intelligence as infrastructure.

Astrix Labs builds foundational systems for AI you can trust — reliable, governed, and traceable. We don't ship clever prompts. We build the substrate the next generation of intelligent systems runs on.

Two productsOne convictionBuilt for accountable teams
01 What we build

Two products, one conviction.

A commercial platform for regulated finance, and the open-source engine that earns developers' trust. Both built on the belief that intelligence should be infrastructure — not a black box.

Flagship platform Private credit & alternative investments

AstroInvestment analysis at AI speed — with an analyst's paper trail.

The governed AI platform for private-credit firms. Astro reads your deal room, drafts IC-grade analysis, and traces every figure back to the exact page it came from.

Ask in plain English; get answers cited to source.
Investment-grade memos and a structured bull/bear Deal Room.
Audit-ready by design — SOC 2-aligned, your data stays in your workspace.
Astro/Meridian Capital/Cascade CIM Lyra ready
What's net leverage at close, and how much covenant headroom does that leave us?
Lyraax-7f2d
Net debt at close is ~$148M against pro forma EBITDA of $34.8M [1] 4.25x net leverage. The senior secured covenant sits at 4.5x [2], leaving ~25bps of headroom.
Drawn from 2 documents[1] Mgmt. Addendum[2] Credit Agreement
genesys — memory server
$ pip install genesys-memory
Installed genesys-memory · Apache 2.0
$ genesys serve --backend obsidian
MCP server up on :8000
11 memory tools registered
score = relevance × connectivity × reactivation
Open source Apache 2.0 · Speaks MCP natively

GenesysThe intelligence layer for AI agent memory.

A scoring engine, causal graph, and lifecycle manager that makes AI memory actually work — so agents recall what matters, forget what doesn't, and can explain why.

89.9% on LoCoMo — ahead of Zep and Mem0.
Drop-in MCP tools for any compliant client.
Runs fully local — Obsidian, Postgres, FalkorDB, or in-memory.
02 What we believe

The challenge isn't model capability. It's orchestration.

As AI moves from experiment to production, the systems that win are the ones teams can trust, inspect, and improve. Four convictions shape everything we build.

01

No black boxes.

Tool access, execution paths, and every answer should be observable, permissioned, and traceable. If you can't see how a system reached a conclusion, you can't trust it.

02

Modularity beats monoliths.

Complex intelligence should be built from small, reusable units that can be composed, versioned, and understood in isolation — not one monolithic prompt.

03

Structure enables scale.

Graphs, constraints, and explicit relationships are what let systems grow without collapsing under their own complexity.

04

Intelligence should improve itself.

The best systems learn from outcomes and feedback, continuously refining how they reason over time.

We are not an AI features company. We are not a prompt library. We build the substrate for the next generation of intelligent systems.

Get started

Build on infrastructure you can trust.

Astro · enterprise platform  ·  Genesys · open source, Apache 2.0