Skip to content

zidane/providers/arcee

The Arcee provider for the Trinity family of open models.

import { … } from 'zidane/providers/arcee'src/providers/arcee.ts

Use arcee() when your agent talks to Arcee models such as Trinity-Large-Thinking. It reads ARCEE_API_KEY when you pass no key. Override baseURL to reach the same models through another host, and use that host's model slug.

TypeScript
import { createAgent } from 'zidane'
import { arcee } from 'zidane/providers/arcee'

const agent = createAgent({ provider: arcee({ defaultModel: 'trinity-large-thinking' }) })
const stats = await agent.run({ prompt: 'Summarize README.md in three bullets.' })
console.log(stats.turns)

Functions#

arceeFunction#

Arcee provider (Trinity-Large-Thinking and friends).

Thin wrapper around openaiCompat pinned to the reasoning_content round-trip mode: Trinity emits chain-of-thought in <think>…</think> blocks that vLLM parses into a plain reasoning_content string. We capture it and echo it back on every assistant message so the model resumes its prior reasoning across tool calls — without this, tool calls leak into the reasoning field as raw XML (Arcee's xml_in_reasoning pitfall). content is also normalized from null → '' on tool-call turns per the same guidance.

See https://docs.arcee.ai/capabilities/reasoning-traces.

TypeScript
function arcee(params?: ArceeParams): Provider
Parameters
NameTypeDescription
params?ArceeParams

API key, base URL, default model, capabilities, reasoning field and extra headers.

Example
import { createAgent } from 'zidane'
import { arcee } from 'zidane/providers/arcee'

const agent = createAgent({
  provider: arcee({ apiKey: process.env.ARCEE_API_KEY }),
})
await agent.run({ prompt: 'Say hello.' })

src/providers/arcee.ts:81

Interfaces#

ArceeParamsInterface#

Options for arcee.

TypeScript
interface ArceeParams
ArceeParams properties
NameTypeDefaultDescription
apiKey?string

API key. Falls back to the ARCEE_API_KEY environment variable.

baseURL?string

Override the base URL — e.g. point at a self-hosted vLLM deployment serving Trinity-Thinking instead of Arcee's hosted Platform API.

capabilities?ProviderCapabilities

Provider capability flags. Trinity models are text-only — default: { vision: false, imageInToolResult: false }.

defaultModel?string'trinity-large-thinking'

Model used when a request does not set one.

extraHeaders?Record<string, string>

Extra HTTP headers sent on every request. Forwarded to the underlying openaiCompat client — use for routing/attribution headers a proxy in front of baseURL consumes.

reasoningContentField?'reasoning_content' | 'reasoning'

Wire field used to echo plain reasoning back across turns. Arcee's hosted API accepts both names; some self-hosted vLLM versions only read reasoning on input (vllm#38488). Default: 'reasoning'.

src/providers/arcee.ts:13