Custom Anthropic tokens; error handling; various bugfixes and refactorings

[berekuk]

Custom token field:

(Doesn't look great, I think it's time to collapse settings in an accordion, but that's out of scope of this PR)

---

Tooltips for failed steps and error messages:

Renamed "Actions" to "Steps", for consistency (I think "Actions" could be confusing for end users, and we already use "Steps" in the sidebar):

Weird small feature: /admin/dev page in dev mode that allows to disable Prisma logs on the running server; doesn't work great but better than nothing when you want to read AI logs and Prisma logs get in the way:

---

LLMClient changes:

• extracted OpenAI-specific and Anthropic-specific code into separate classes
• implemented LLMError class that allows to distinguish between "timeout" errors and "out of credits" errors

Error handling:

• queryLLM now always returns a completion, or fails when something went wrong
• more errors are critical, afaict it shouldn't prevent any runs that could succeed on retries

Streaming, ClientWorkflow and events refactorings:

• I've removed stepStarted event, we update the DB on stepAdded events now (minor difference but fixes the issue with in-progress status)
• also removed clientWorkflow.currentStep, could be inferred from the last step, and this fixes one bug where we showed step id instead of step name in the sidebar

[changeset-bot]

⚠️ No Changeset found

Latest commit: 9323a7a

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
quri-hub	✅ Ready (Inspect)	Visit Preview	Nov 25, 2024 9:49pm
squiggle-website	✅ Ready (Inspect)	Visit Preview	Nov 25, 2024 9:49pm

2 Skipped Deployments

Name	Status	Preview	Updated (UTC)
quri-ui	⬜️ Ignored (Inspect)	Visit Preview	Nov 25, 2024 9:49pm
squiggle-components	⬜️ Ignored (Inspect)	Visit Preview	Nov 25, 2024 9:49pm

[OAGr]

This is failing tests

[OAGr]

Reasonable move, to split this up.

[OAGr]

That said, this does make it tricky to determine what code was added / changed, of this part.

[berekuk]

IIRC only LLMError class is new, everything else in LLMClient/ is old.

And even that class is used only for timeout errors (specific LLM providers don't try to normalize vendor lib errors to LLMError instances, which works because everything that's not an LLMError is treated as critical, which is what we want, for now).

[berekuk]

This is failing tests

Oops, yeah, that's because of f84baa1 that I noticed and fixed but didn't backport here (Vercel started to default to Node v22, probably recently).