Tutorials and source examples

Start with a local prompt, then add the capabilities your task needs. You do not need a daemon, shell access, or an orchestration script to try Ochat.

Learn in order

1. Run your first local agent: setup, one request, and clean exit.
2. Give it a file tool: a complete read declaration and a named sample file.
3. Add a specialist reviewer: a parent, companion prompt, and observable delegation.
4. Run a request from a script: initialize and continue a file-backed conversation.
5. Add bounded ChatML workflow logic: stop after three completed turns.
6. Give it a narrow shell command: inspect authority and select a supported host.
7. Run a durable daemon session: detach, reconnect, and shut down.
8. Respond to a background event: a timer without model work.
9. Connect a stdio client: initialization, attachment, and streamed completion.
10. Connect an HTTP client: authenticated RPCs and event subscriptions.

Each tutorial records its host, prerequisites, verification scope, and persistence. Read the associated ChatMD and companion files directly in each tutorial’s source reader or in the website catalog. Download a bundle when you are ready to run it in your configured local environment. The website does not execute agents in your browser.

Choosing a source example

Complete examples include their local source/data dependencies; Ochat installation and any stated provider credentials remain prerequisites. Configurable templates need the additional host, backend, library, or connection setup described in their linked tutorial. Illustrative output is a reading sample, not executable input.

On the website, expand View source in the catalog to read complete files with syntax highlighting. The entrypoint is shown first; expand other filenames to inspect companion prompts, ChatML scripts, sample data, build files, and notices. The reader also works without JavaScript. Downloads provide exact-byte source files and complete .tar bundles. Extract a bundle into a new directory with tar -xf FILE.tar; its top-level directory matches the example ID. Preserve its relative paths and license. Download every listed companion or use the bundle before running a multi-file example. In a checkout, the following links open maintained sources directly.

Add useful capabilities

The repository-only longer prompt examples preserve the minimal, refactoring, and moderator examples from the detailed project README, including the Item, Tool_call, and Context helper walkthrough. They remain deferred from website publication pending example-level review; use the complete tutorials above for the supported learning progression.

Try this	Example or walkthrough
Read project files or add editing tools	Tool declarations and catalog
Compose a larger assistant from prompts and tools	General agent workflow
Search code and documentation	Search setup and illustrative output samples
Give an agent a narrow command	Complete shell tutorial
Explore other shell configurations	Shell declaration patterns
Respond to a scheduled background event	ChatML background agent
Implement a custom OCaml tool with progress	Tool registration guide and compiled offline example

The prompt example collection contains larger prompts to study and adapt. Treat these and the workflow examples as starting points, not a promise that their paths, services, models, or permissions match your environment. View ChatMD as source/code so its tags remain visible.

For an actual run rather than a template, see the documentation-update session, which links both the full transcript and its compacted version. These historical artifacts illustrate the workflow; they are not current setup instructions.

Host an agent or build a client

• Unix daemon and TUI: create a session and reconnect to it independently of the terminal client.
• Stdio client: use the Ochat protocol through a local host or daemon gateway.
• HTTP client: authenticate requests and consume live updates.
• Tracked server examples: reusable prompts, client inputs, and a helper that generates private demonstration configuration.
• Example configuration notes: where generated configurations and credentials belong.

The tracked examples include a tool-free prompt, a timer script, and a narrow shell prompt. Use the accompanying tutorials for setup rather than assuming each file runs without configuration.

Before running an example

Check the host mode, provider configuration, tool access, and required services. Live model calls may cost money; generated demo credentials are for isolated local demonstrations, not a production authentication policy. Use a disposable workspace when trying tools that can change files or execute commands.

For explanations behind the examples, see ChatMD, tools, ChatML, and agent hosting. The documentation home links the full reference set.