Agentic tool settings

Agentic tools send file contents, command output, and the whole conversation history at every step. A prompt that large can take tens of seconds before the first token arrives. If your app gives up first, the request still counts but you receive nothing. Set timeout, streaming, retries, and context size as shown below.

Why this matters

• Requests cut off before the response arrives still count. As soon as a request arrives, the AI server starts processing the whole prompt. If your app drops it first (timeout, stop button, lost connection), the processing cost has already been incurred, but you get no response.
• Huge prompts need more than 30 seconds before the first token. Agentic prompts of hundreds of KB with thousands of messages can exceed many apps’ default limit (30 or 60 seconds) before a single token comes out.
• Streaming keeps the connection alive. With streaming on, our server sends the headers right away and then a keepalive signal every 10 seconds while waiting for the response. What can still cut it off is your app’s total time limit, so set it to at least 300 seconds.
• Pick a fast model for large contexts. For long sessions or big repos, use a faster model and keep large models for the steps that really need them. Compare them on the Pricing page.

Guidelines for every tool

Timeout

At least 300 seconds. Watch the unit: some tools use seconds (300), others milliseconds (300000).

Streaming

Turn it on. Without streaming, no data flows until the response is complete, so proxies and apps give up sooner.

Retries

Keep them low (0–3). A retry after a timeout is a new request that is also charged. Fix the timeout first instead of raising retries.

Context

Keep it small. Use automatic compact/summarize, start a new session per task, and exclude large folders (node_modules, build, data).

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://ai.geraikita.com",
    "ANTHROPIC_AUTH_TOKEN": "gk-xxxxxxxx",
    "API_TIMEOUT_MS": "600000",
    "CLAUDE_CODE_MAX_RETRIES": "3",
    "CLAUDE_AUTOCOMPACT_PCT_OVERRIDE": "60"
  }
}

{
  "provider": {
    "geraikita": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Geraikita AI",
      "options": {
        "baseURL": "https://ai.geraikita.com/v1",
        "apiKey": "gk-xxxxxxxx",
        "timeout": 600000
      },
      "models": {
        "claude-sonnet-5": {
          "name": "Claude Sonnet 5",
          "limit": { "context": 128000, "output": 16000 }
        }
      }
    }
  },
  "compaction": { "auto": true }
}

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "geraikita": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://ai.geraikita.com/v1",
        "apiKey": "{env:GERAIKITA_API_KEY}",
        "headerTimeout": 300000,
        "chunkTimeout": 300000
      }
    }
  },
  "compaction": { "auto": true }
}

models:
  - name: Claude Sonnet 5 (Geraikita)
    provider: openai
    model: claude-sonnet-5
    apiBase: https://ai.geraikita.com/v1
    apiKey: gk-xxxxxxxx
    requestOptions:
      timeout: 600        # seconds, not milliseconds
    defaultCompletionOptions:
      contextLength: 64000  # smaller = history gets trimmed
      maxTokens: 8192

from openai import OpenAI

client = OpenAI(
    base_url="https://ai.geraikita.com/v1",
    api_key="gk-xxxxxxxx",
    timeout=600.0,   # seconds
    max_retries=1,   # retries after a timeout are charged too
)
stream = client.chat.completions.create(
    model="claude-sonnet-5",
    messages=[{"role": "user", "content": "Halo!"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

import anthropic

client = anthropic.Anthropic(
    base_url="https://ai.geraikita.com",   # without /v1
    api_key="gk-xxxxxxxx",
    timeout=600.0,   # seconds
    max_retries=1,   # retries after a timeout are charged too
)
with client.messages.stream(
    model="claude-sonnet-5",
    max_tokens=4096,
    messages=[{"role": "user", "content": "Halo!"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

curl https://ai.geraikita.com/v1/chat/completions \
  --max-time 600 -N \
  -H "Authorization: Bearer gk-xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-5","stream":true,"messages":[{"role":"user","content":"Halo!"}]}'

Billing terms

FAQ