Explore Agentic Kit / Autonomous AI Stack

Templates that run, and a model that arrives trained.

The kit is the part that turns a callable surface into something that works while you are not watching: ninety forkable crates, the runtime that keeps them alive, and a supervising model you configure rather than train.

Pick the AI-model

model amodel bmodel c
Supervisinglive
model amodel bmodel c
Strategyspot-grid

The model above changes. This does not.

In numbers

What you are starting from.

90
Crates in the repository

agent-* in silvana-book-agent

68
Agents in the catalogue

app.silvana.one/api/agents

6
Families, execution to compliance

AGENT_FAMILIES

4
Archetypes doing the work

AGENT_ARCHETYPES

What is in the kit

Six families, ninety crates, none of them a snippet.

Whatever the agent is for, something in here already does most of it. The families are the aisle; the crates are the shelf.

execution

Execution

Work an order without moving the market against yourself.

  • TWAP Execution
  • Iceberg Execution
  • Algo Order
  • Liquidity Seeking
strategy

Strategy

Take a position on a rule you can state and test.

  • Spot Grid
  • Trend Following
  • Pairs Trading
  • Mean Reversion
  • Arbitrage Agent
portfolio

Portfolio

Hold a shape across instruments instead of one trade at a time.

  • Portfolio Rebalancing
  • Target Allocation
  • Smart Allocation
  • DCA Portfolio
risk

Risk and control

The agents that stop the other agents.

  • Killswitch
  • Circuit Breaker
  • Risk Management
  • Pre-Trade Check
  • Concentration Risk
oracle

Oracle and data

Bring outside truth on-chain, and publish what happened.

  • Oracle
  • Fair Value
  • Orderbook Streaming
  • Volatility Screening
compliance

Compliance and audit

Prove the run, not just the result.

  • Audit Attestation
  • Audit Replay
  • Compliance Screening
  • Selective Disclosure
What is doing the work

Four kinds of thing can hold a job.

Independent of the family above — a portfolio job can be plain automation or a model in an enclave. This is the axis that decides how much of the runtime you are relying on.

automation.rs

Automation agents

Sequence transactions and app operations — request handling, account abstraction, flow automation, and scheduled, pre-paid, delayed or event-triggered transactions.

Deterministic. Given the same inputs it does the same thing, every time.

oracle.rs

Oracle and witness agents

Witness and fetch on- and off-chain business data, and publish it provably so the rest of the workflow can rely on it.

Brings in facts from outside the ledger, with a proof attached.

ai.rs

AI agents

Run inside secure enclaves to produce insights and encrypted interpretations of data the raw numbers do not state outright.

Judgement. This is the archetype that can decide the rule no longer applies.

human.rs

Human agents

Professionals and individuals providing manual checks, audits or expert opinion at the points a workflow calls for one.

A person, addressed through the same interface as the code.

Autonomous AI stack

Configure the model. Do not go and train one.

The slowest part of putting a model over a strategy is usually everything that happens before the strategy: choosing one, wiring it, training it, evaluating it. This is the part Silvana supplies.

model-as-component.rs

The model is a component, not a fixture

Swap it, run two and compare them, or combine them. The strategy underneath does not change when you do — which is what makes the choice reversible rather than a commitment made on day one.

arrives-trained.rs

It arrives trained for the job

You are not standing up an ML practice to get an agent running. Silvana supplies the model attached to the agent and trained for the tasks the catalogue covers, so the work in front of you is your logic.

teach-it-yours.rs

Then teach it how you trade

Your limits, your instruments, your idea of an acceptable fill. The point of a supervising model is that it learns the strategy you already run rather than imposing a generic one.

enclave.rs

Somewhere you can reason about

The AI archetype runs inside secure enclaves, producing encrypted interpretations of data rather than shipping the raw numbers somewhere. That is the archetype with the authority to decide a rule has stopped applying.

The agents are open source. Every claim above is something you can check in the crates before you trust it with a position.

Try one first

Open a running agent.

Every catalogue entry has a demo you can open and a repository you can read. No signup between you and either.

catalogue

Sixty-eight agents in the catalogue, ninety crates behind them. Deploy one from the app in minutes, or fork the closest crate and make it yours.

Start

Fork the closest one.

The quickstart puts a configured agent on the network. After that, the difference between a template and your agent is the logic you were going to write anyway.