Skip to main content

Arcee Provider Guide

Arcee is a Tier-2 catalog provider: OpenAI-wire-compatible, so its entire integration is one JSON file (src/lib/providers/catalog/arcee.json) rather than hand-written code. That file is the source of truth for everything on this page.

Verification status: this entry is docs-verified only, not yet live-verified. The vendor's GET https://api.arcee.ai/api/v1/models needs a key (it answers HTTP 401 without one), so the model ids come from Arcee's public docs and the roster has not been checked. No account was created and no API key was used to build this entry. evidence.liveMatrix is null until someone runs the live capability matrix with a real key (see Verification status below).

Key Facts​

  • Provider id: arcee
  • Protocol: OpenAI-compatible (/chat/completions). The Your First API Call page says "The Arcee Platform API is OpenAI compatible."
  • Base URL: https://api.arcee.ai/api/v1 — the base_url value in the Quick Start Python example and the baseURL value in its JavaScript example. The Your First API Call page table lists base_url as https://api.arcee.ai/api/v1/chat/completions; wire.baseURL is the value the SDK examples set
  • Default model: trinity-large-thinking — the API Model Name the Arcee Model Overview lists for Trinity-Large-Thinking (400B), which the page says is "the only Arcee model available via api."
  • Models in catalog: 10 — the ten rows of the "Text Models" table on the pricing page
  • Streaming: supported — the Streaming Messages page says "To enable streaming, set stream=True."
  • Tool calling: model-dependent — the Function Calling page documents the tools parameter and the tool_calls output with a MODEL_NAME placeholder, and the Trinity-Large-Thinking page lists "Purpose-built for tool calling, multi-step planning, and agent workflows" under Key Features
  • Tools while streaming: false — not exercised, since no POST request was sent
  • Structured output: true — the Structured Outputs page documents the response_format value {'type': 'json_object'}. Under "To enable JSON Output, users should:" the page lists "Include the word "json" in the system or user prompt, and provide an example of the desired JSON format to guide the model in outputting valid JSON." and "Set the max_tokens parameter reasonably to prevent the JSON string from being truncated midway." quirks.responseFormatDowngrade is set so a generate({ schema }) request sends json_object; NeuroLink still validates the response against your schema on the client. No request with json_schema was sent
  • Structured output + tools together: false — no combined probe was run without credentials
  • Embeddings: false
  • Thinking: false — the entry sets no thinking-level request parameter. The Reasoning Traces page documents the reasoning_content response field for Trinity-Large-Thinking
  • Reasoning replay (quirks.replayReasoningContent: true): the Trinity-Large-Thinking page says "thinking tokens must be kept in context for multi-turn conversations and agentic loops to function correctly", and the Reasoning Traces page says "use reasoning_content when constructing assistant messages for input". NeuroLink sends each assistant turn's reasoning_content back on later requests. No multi-turn request was sent
  • Billing: schema value no-free-tier, see Billing below
  • Key format: none declared. The Chat Completion reference gives the key format as "Arcee API key (prefixed with rcai-)"; NeuroLink does not validate it because no key was available to test it

Quick Start​

1. Get an API key​

  1. Visit: https://platform.arcee.ai/api/api-keys — the Quick Start says to access the Arcee Platform (https://platform.arcee.ai/), "Register or Login.", then to "Create an API Key in the API Keys management page." and "Copy your API Key for use and store in a secure location."
  2. Authentication is HTTP Bearer: the Quick Start's curl example sends the header Authorization: Bearer YOUR_API_KEY, and the Chat Completion reference (https://docs.arcee.ai/api-reference/chat-completion.md) gives the key format as "Arcee API key (prefixed with rcai-)"
  3. Billing as the vendor states it: the Quick Start says to open the Wallet (https://platform.arcee.ai/api/wallet) "to top up if needed"; the Chat Completion reference lists HTTP 402 as "Insufficient Balance"
  4. Set ARCEE_API_KEY in your .env file

2. Configure​

export ARCEE_API_KEY=your-api-key
export ARCEE_MODEL=trinity-large-thinking   # optional — overrides the default model
export ARCEE_BASE_URL=https://api.arcee.ai/api/v1   # optional — proxy or self-hosted gateway

3. Use it​

import { NeuroLink } from "@juspay/neurolink";

const neurolink = new NeuroLink();

const result = await neurolink.generate({
  input: { text: "Explain context windows in one paragraph." },
  provider: "arcee",
  model: "trinity-large-thinking",
});

console.log(result.content);
# CLI
npx @juspay/neurolink generate "Hello" --provider arcee

Per-request credentials use the same shape as the other catalog providers:

await neurolink.generate({
  input: { text: "Hello" },
  provider: "arcee",
  credentials: { arcee: { apiKey: process.env.ARCEE_API_KEY } },
});

Billing​

The Quick Start says to open the Wallet (https://platform.arcee.ai/api/wallet) "to top up if needed". The Chat Completion reference lists HTTP 402 as "Insufficient Balance". The pricing page says "All prices are in USD." and lists prices per 1M tokens.

The catalog schema has no unknown billing value, so billingPolicy holds no-free-tier. It is a schema value, not a vendor statement of signup terms.

Models​

ModelContextVision$/M in · out (cached)Notes
trinity-large-thinking ⭐128,000no$0.25 / $0.80 ($0.06)NeuroLink default. Strength "Robust generalist model with strong performance across reasoning, coding, math, and complex task decomposition." Context Window "512k tokens (hosted at 128k)".
deepseek/deepseek-v4-flash-latest—no$0.14 / $0.28 ($0.028)NeuroLink fallback; listed under "Text Models".
zai-org/glm-5.3-flash—no$0.15 / $0.50 ($0.03)NeuroLink fallback; listed under "Text Models".
thinkingmachines/inkling-small—no$0.50 / $1.20 ($0.10)NeuroLink fallback; listed under "Text Models".
deepseek/deepseek-v4-pro-0813—no$1.32 / $3.96 ($0.044)NeuroLink fallback; listed under "Text Models".
deepseek/deepseek-v4-pro—no$1.74 / $3.48 ($0.20)NeuroLink fallback; listed under "Text Models".
zai-org/glm-5.2—no$1.40 / $4.40 ($0.26)NeuroLink fallback; listed under "Text Models".
zai-org/glm-5.3—no$1.40 / $4.40 ($0.26)NeuroLink fallback; listed under "Text Models".
moonshotai/kimi-k3—no$3.00 / $15.00 ($0.30)NeuroLink fallback; listed under "Text Models".
deepseek/deepseek-v4.1-flash—no$0.22 / $0.66 ($0.007)NeuroLink fallback; listed under "Text Models".

Model ids and prices are copied from the pricing page, retrieved 2026-09-29; the page says "This page provides pricing information for all models on Arcee Platform." and "Prices per 1M Tokens." The default's id, Context Window and Strength come from the Arcee Model Overview, retrieved 2026-09-29. A — in the Context column means the catalog carries no contextWindow for that model, so it uses models.defaultContextWindow.

On www.arcee.ai/trinity the card titled "Trinity Large Thinking (400B, 13B active)" gives "Arcee API — served at 256K context at BF16" after its "Deploy on:" label. The Arcee Model Overview gives "512k tokens (hosted at 128k)" for the same model. The two vendor pages give different hosted context figures; the catalog's contextWindow for trinity-large-thinking is 128000, the Model Overview figure, the page that lists the API Model Name.

The Vision column shows the vision value in the JSON, false on the ten rows. It is a placeholder for a boolean the catalog schema requires; it is not a vendor statement.

Each model's status in the JSON is production, a placeholder for a value the catalog schema requires; it is not a vendor statement.

models.defaultContextWindow (128000) and models.defaultMaxOutputTokens (8192) apply to model ids outside the catalog and to catalog rows without their own contextWindow or maxOutputTokens (nine rows have no contextWindow; ten rows have no maxOutputTokens). They are placeholders the vendor does not publish for those model ids; 128000 is the same number as the "hosted at 128k" figure the Arcee Model Overview gives for trinity-large-thinking.

Fallback order when the default is unavailable follows the row order of the pricing page: deepseek/deepseek-v4-flash-latest → zai-org/glm-5.3-flash → thinkingmachines/inkling-small → deepseek/deepseek-v4-pro-0813 → deepseek/deepseek-v4-pro → zai-org/glm-5.2 → zai-org/glm-5.3 → moonshotai/kimi-k3 → deepseek/deepseek-v4.1-flash. The runtime fallback model name the loader derives (fallbacks[1]) is zai-org/glm-5.3-flash.

Model changes. The Deprecation Policy says "Cloud models are versioned and may be replaced quickly when new improvements are available." and "Typical notice for standard cloud models: up to 5 business days".

Verification status​

Tier-2 onboarding requires evidence before a provider is accepted, and pnpm run verify:provider-onboarding gates it in CI. This is what the catalog currently records for Arcee — docs-verified only, not live-verified:

ProbeResult
Rosterunauthenticated GET /api/v1/models answers HTTP 401 without a key (2026-09-29), so model ids are taken from the pricing page; roster not verified. The Models reference documents this endpoint with a 401 response described as "Authentication Fails"
Auth-failure shapeunauthenticated GET /models answered HTTP 401 with header www-authenticate: Bearer and body {"detail":"Missing or invalid Authorization header. Expected: Bearer <api_key>"} (2026-09-29); the Chat Completion reference Error schema has one property, detail. errorRules carry the 401, 402, 403, 429 and 503 rows
/chat/completions GETunauthenticated GET /chat/completions answered HTTP 405 with header allow: POST and body {"error":{"message":"Method Not Allowed","type":"invalid_request_error","param":null,"code":"request.rejected"}} (2026-09-29). No POST request was sent
Billingpublic pages quoted under Billing; no signup performed
Tools / structured outputdocumented on the Function Calling and Structured Outputs pages; neither was exercised live, no POST request was sent, and no combined tools+schema request was sent, so structuredOutputWithTools stays false. The request fields NeuroLink's chat-completions client can add have not been exercised against Arcee
Live capability sweepnot run. evidence.liveMatrix is null. Before treating this provider as production-ready, run npx tsx test/continuous-test-suite-provider-matrix.ts --provider=arcee with a real key and record the result.

Do not treat this entry as equivalent to a live-verified Tier-2 provider (e.g. FriendliAI, Novita AI) until that live matrix has been run and evidence.liveMatrix is filled in.

Troubleshooting​

The rows below are the statuses the vendor's Chat Completion reference lists; NeuroLink's errorRules map 401, 402, 403, 429 and 503 to error classes.

SymptomWhat the Chat Completion reference says
HTTP 400"Invalid Format"
HTTP 401"Authentication Fails"
HTTP 402"Insufficient Balance"
HTTP 403"Permission Denied"
HTTP 404"Not Found"
HTTP 422"Invalid Parameters"
HTTP 429"Rate Limit Reached"
HTTP 500"Server Error"
HTTP 503"Server Overloaded"

See also​