Discord
Real-time feedback, support requests, integration debugging, and community coordination.
Join @agentralabsDOMAIN AI THAT REASONS.
INFRASTRUCTURE THAT REMEMBERS.
Your AI guesses. Ours reasons like a 30-year domain expert — and remembers every decision it ever made.
Three domain-specialist models. Open-source infrastructure. Every decision verifiable.
SOLEN · VERAC · AXIOM
quickstart.terminal
solen.global
>ollama run agentralabs/solen-e4b
Run Solen locally — supply chain reasoning.
>pip install solen-sdk
Python SDK for Solen inference.
// SECTION: THE_STACK002
architecture.overview3 layers · 1 stack
The only stack where a domain-specialist AI can make decisions a regulator can audit.
Reasoning → Substrate → Verification
// SECTION: MODELS003
live surface: 3 domain-specialist models · active operations
MODELS
IN TRAINING01/03
Solen / SUPPLY CHAIN MANAGEMENT
Thinks like a supply chain director. Not a chatbot that knows about supply chain. Solen has read nothing but supply chain its entire existence — every research paper, every SEC filing, every court case where a supply chain failure ended in litigation.
—
Clarification-first reasoning. Tells you exactly what it needs to know and why, before it answers. Reasons through incomplete data with calibrated confidence.
—
Domain saturation. Trained on nothing but supply chain — every research paper, every SEC filing, every court case where a supply chain failure ended in litigation.
—
Powers Nexus Planner. The reasoning engine behind Nexus Planner, Agentra Labs' supply chain intelligence surface.
MODULE.GLYPH
MODELS
IN TRAINING02/03
Verac / FINANCE / SETTLEMENT OPERATIONS
Every invoice, every reconciliation, every settlement decision stays inside your infrastructure. Verac runs on your servers. Nothing leaves. Data residency by design.
—
On-premise by design. Every invoice, every reconciliation, every settlement decision stays inside your infrastructure. Nothing leaves. Data residency is not optional.
—
Bank-grade trust model. On-premise financial reasoning. The condition under which bank adoption becomes possible.
—
Powers ZexRail. The reasoning engine behind ZexRail, multi-rail payment settlement with real-time reconciliation.
MODULE.GLYPH
MODELS
IN TRAINING03/03
Axiom / FINANCIAL MARKETS / TRADING
Financial markets reasoning with calibrated confidence. Consequence reasoning — reasons forward to second and third order effects before recommending action.
—
Consequence reasoning. Reasons forward to second and third order effects before recommending action. Does not chase correlations.
—
Missing-data awareness. Identifies what is missing before answering. Calibrated confidence over false precision.
—
Standalone deployment. Runs independently of the Agentra product stack. Plug into any trading infrastructure via standard APIs.
MODULE.GLYPH
The Specialist Argument
Nobody hires a generalist lawyer to argue a patent case.
These models are trained on reasoning processes, not questions and answers. Domain saturation produces specialists that think inside a field — not chatbots that have read about it.
VerificationIncomplete DataError DetectionExpert ConflictConsequence ReasoningFailure Pattern Recognition
// SECTION: SUBSTRATE_FLAGSHIP004
cognitive.graphlive
Every decision traces back to the facts that caused it. Corrections link to what they replaced. Truth has a history.
MEMORY
Most AI memory is retrieval over flattened text. AgenticMemory is graph cognition — nodes for what the agent learned, edges for why things connect, traversal for reasoning history.
16 query types. 6 event types. One portable .amem file per agent brain. No cloud database. No vector service. Memory-mappable, offline-capable, and designed for 20 years.
MCP Tools
0
Query Types
0
Format
.amem
How Memory Serves the Models
STEP 01
Solen recommends changing suppliers
STEP 02
Memory stores the reasoning chain — 3 facts, 2 decisions, 1 inference
STEP 03
6 months later: "why did we switch?" — the chain is intact
Nothing was forgotten. Nothing was hallucinated.
// SECTION: SUBSTRATE_DIRECTORY005
Open-Source Substrate
18 OPEN-SOURCE
SYSTEMS
The substrate your models run on
MODULE.GLYPH
0+MCP tools
0file formats
0MIT repos
Substrate Directory18 MIT
#ProjectArtifactTools
STANDARD
All open source. Published on crates.io, PyPI, npm.
// SECTION: PROOF006
Model Benchmarks
1Solen vs GPT-4o (Supply Chain)
Coming soon2Verac vs GPT-4o (Finance)
Coming soon3Axiom vs GPT-4o (Markets)
Coming soonGraded by Claude (neutral). Methodology published in full.
Substrate Speed
Memory: causal traversal< 1 ms
Memory: semantic search (100K nodes)< 10 ms
Codebase: symbol lookup14.3 µs
All MCP p99< 100 ms
MODULE.GLYPH
LOCAL-FIRSTEvery model runs on your hardware. Every .amem file lives on your disk. Zero cloud dependency.
MODULE.GLYPH
DATA STAYS ON-PREMISEVerac runs on your servers. Nothing leaves.
MODULE.GLYPH
20-YEAR MEMORY2 GB of reasoning continuity in a single .amem file
// SECTION: SHOWCASE007
HYDRA.NETWORK
HydraShowcase
The living proof that the stack composes.
68 Rust crates. Persistent memory via AgenticMemory. Self-writing genome. Constitutional governance via AgenticContract. Planning via AgenticPlanning.
When Hydra needs to reason about supply chain, it calls Solen. When it needs finance, it calls Verac. The first customer of the entire stack.
68CRATES
SELF-WRITINGGENOME
CONSTITUTIONALGOVERNANCE
Release Broadcast
live: github releases
// SECTION: COLLABORATION_CTA003B
Built for teams that need auditable AI decisions in regulated industries
We collaborate with research labs, enterprise engineering teams, and infrastructure sponsors. Our stack combines domain-specialist models for reasoning, open-source substrate for memory and governance, and deterministic verification for every decision.
// SECTION: ROUTING_TABLE017
Channel routing
X (Twitter)
Showcase highlights, launch updates, and public signal from builders.
Follow @agentralabFormal contact for integrations, support escalation, and collaboration requests.
contact@agentralabs.tech