A trading agent on Canton, without the ledger work.
Rust crates and a CLI. The communication layer is abstracted, signing stays on your side, and ninety readable crates already do the parts that are not your edge.
With the SDK
// the track is already bolted down, in order
- ledger client
- message signing
- order lifecycle
- reconnection
- settlement watcher
On your own
// same distance, and you lay every one of those yourself
Both reach Canton. One of them is a week you do not spend on your own edge.
Four things you would otherwise write first.
None of them are your strategy, and all of them have to work before your strategy gets to run.
You never touch the ledger layer
The docs put it plainly: the communication layer is abstracted and what you get is an interface to the orderbook and settlement flows. No Daml, no contract lifecycle, no ledger client of your own.
Signing stays local
Ed25519 message-level signing happens on your side. Nothing in the flow asks you to hand a key to a service, which is the difference between an integration and a custody decision.
Configuration instead of a deployment story
Market, spread, size, levels, limits and strategy behaviour are settings in a file. The CLI runs it. That is the whole operational surface.
Ninety starting points
Order lifecycle, reconnection, settlement watching — written, and written ninety times over in crates you can read. Fork the closest one instead of rediscovering the same edge cases.
What is already there.
- 90
- Crates to fork
- 18
- SDK doc pages
- 5
- Book services on gRPC
- 68
- Agents in the catalogue
agent-* in silvana-book-agent
/sdk/ in docs.silvana.one
execution, pricing, settlement, ledger, news
app.silvana.one/api/agents
The path, in order.
Four documents, taken in this sequence, end with an agent placing a live order.
The shortest complete route. Everything further down is a detour off it.
- PrerequisitesToolchain, keys and accounts to have in place before anything below will run.
- SDK quickstartEmpty directory to an agent placing orders, start to finish.
- Agent onboardingRegistering an agent identity and getting it authorised to trade.
- Agent sampleThe smallest complete agent, annotated — the thing to copy from.
// test-account
Ask for a funded test account before you risk anything
Silvana issues test accounts for QA and working promo codes for developers. Neither is self-serve yet — they come from a person, so the contact form is the whole process.
- Which environment you want — devnet or the test book
- What you are building, in a sentence
- Whether you need a promo code as well as an account
// net change after every round trip: 0.00
Test funds settle the same way real ones do. The difference is what a mistake costs you.
The rest of the surface.
Building and configuring an agent.
- SDK introductionWhat the SDK covers and the shape of an agent built on it.
- Silvana Book agentThe trading layer: orders, fills, the book itself.
- ConfigurationEvery setting an agent reads from agent.toml, and its default.
- Full CLI referenceEvery command, flag and subcommand.
- Trading modesHow an agent places into the book, and where RFQ fits.
Who can sign what, and where it runs.
The fastest way to judge an SDK.
Open an agent built on it. Every catalogue entry ships a crate, a doc.md and a demo that needs no signup.
// 90 crates — strategy, execution, portfolio, risk, oracle, compliance
Grid, TWAP, iceberg, pairs trading, rebalancing, killswitch, circuit breaker, audit replay. Each is a working agent rather than a snippet, and each one you read teaches the same SDK.
Or go straight at the ledger.
If you are connecting a system that already exists rather than building an agent, the gRPC surfaces are the shorter road. There are two of them, and picking takes about a minute.