Orchard fits anywhere a program needs to make a decision with a model and act on it. Here is a sampling of what people build, from a single typed call to a long running agent with its own memory.
Assistants and chat agents
The most direct use is an agent that holds a conversation and gets things done inside it.
- Support assistants that read a ticket, decide what it needs, and draft a reply.
- Internal helpers that answer questions about your own systems and run the lookups for you.
- Product copilots that live inside an app and call its functions as tools.
Classification and routing
When the job is to read something and decide where it goes, typed generation does the heavy lifting. You declare an enum, ask for it, and branch on a result the compiler proves you have handled completely.
- Triage incoming reports by severity and send each to the right queue.
- Tag and route email, tickets, or messages without writing a parser.
- Score content against a policy and return a clear, checked verdict.
Structured extraction
Turn unstructured text into typed records that the rest of your program can rely on.
- Pull fields out of invoices, contracts, or forms into a declared type.
- Read a long document and return a summary object with named parts.
- Normalise messy input into a clean shape before it touches a database.
Tool-using workflows
Give an agent a set of tools and let it decide how to use them. The runtime drives the loop and keeps it inside the budgets you set.
- Agents that call your HTTP APIs, read the web, or run shell commands through the built in packs.
- Workflows that connect to any server speaking the Model Context Protocol.
- Multi step jobs that gather data, act on it, and report the result.
Research and retrieval
Combine generation with memory and search to answer questions that need more than one step.
- Agents that look something up, read the results, and decide what to do next.
- Recall over past conversations using the built in memory store.
- Pipelines that fan work out across sources and bring the findings back together.
Automation and scheduled jobs
Agents do not have to wait for a person. Orchard can start work on a schedule or in response to a change.
- Run an agent on a cron style schedule to summarise, check, or report.
- React when a file changes and process it automatically.
- Serve an agent over HTTP so other systems can call it.
Intelligence embedded in apps
Because the core is a portable library, the same agent can live inside software you already ship.
- Call Orchard from a Rust service or a C program through the stable header.
- Use it from Python with the native bindings.
- Run it in the browser through WebAssembly, with no server in the loop.
Start building
- The progressive guide walks from a one line agent to a tool using one, all runnable offline.
- See why Orchard makes these patterns short and safe to write.