Skip to content
plugboard.appRequest a demo

Connectors

Anthropic (Claude)

Run the service desk’s AI features on Claude, on your school’s own API key and your school’s own account.

CategoryAI
AuthenticationAPI key
Reachesapi.anthropic.com
Needs an agentNo
Demo modeNo

ai.chat, which powers the AI features: Ask, ticket triage, and the summary on a ticket.

You want the best answers and you are content for ticket text to go to Anthropic under your own commercial terms.

The bundled local model is private and free, but slow. This is faster and better, and billed to you. See choosing a model for the comparison.

  1. Create an account at console.anthropic.com.
  2. Add billing. There is no free tier for API use.
  3. API keys, then create one. Copy it; it starts sk-ant-.

Create the key under a workspace for the school rather than a personal one, so it survives whoever set it up moving on, and so the spend is attributable.

Settings, Connectors, Anthropic (Claude), Configure.

FieldDefaultValue
modelemptyOne model id for every task. Empty uses claude-haiku-5-5 for quick tasks and claude-sonnet-5-5 for more demanding ones. claude-opus-5 gives the best answers at the highest cost
maxTokens4096The most tokens one answer may use, thinking included
timeoutMs30000Give up on an answer after this long
FieldValue
apiKeyYour key, starting sk-ant-

Save and test.

Structured answers such as ticket summaries use Anthropic’s schema format. Plugboard validates the returned answer before using it.

Billed by Anthropic per token, on your account. The work this product asks for is small: a triage or a ticket summary is a few thousand tokens in and a few hundred out.

A desk triaging around three hundred tickets a month, plus questions in Ask, lands somewhere between five hundred and fifteen hundred calls. On claude-haiku-5-5 that is under a US dollar a month at the per-token prices Plugboard records for it. Per token, claude-sonnet-5-5 costs about twenty times as much and claude-opus-5 about fifty times as much, and the answers are better.

Start with model empty and set a single model only if the answers are not good enough.

To Anthropic, under your agreement with them rather than ours.

If your privacy assessment needs the data to stay on your own hardware, use the bundled local model instead. The compliance page states which one you have configured.

SymptomCause
401 on testWrong key, or it was revoked in the console
400 about creditBilling is not set up on the account
429Rate limited. Anthropic’s limits rise with account history
Answers cut off mid-sentencemaxTokens is too low for the model you picked
TimeoutsRaise timeoutMs, or pick a faster model
404 on the model idThe id is wrong, or your account has no access to it

OpenAI-compatible endpoint for anything else, including Azure OpenAI and self-hosted GPU boxes. Ollama for the model that ships in the box.