Packages

eai

1.0.2
1.0.6 1.0.5 1.0.4 retired 1.0.3 1.0.2 1.0.1 1.0.0 0.1.15 retired 0.1.14 retired 0.1.11 retired 0.1.9 retired 0.1.8 retired 0.1.7 retired 0.1.6 retired 0.1.5 retired 0.1.4 retired 0.1.3 retired 0.1.2 retired 0.1.1 retired 0.1.0 retired

Extreme minimal AI assistant with persistent PTY, recursive sub-agents, and CLI-based MCP/browser integration (mcporter, agent-browser)

HexDocs

Current section

Files

Jump to
Readme
20 Versions
9 Dependencies
Dependants
Files
Activity
eai config models claude_opus.exs
Raw

config/models/claude_opus.exs 

import Config



config :eai, :model_claude_opus,

  name: :claude_opus,

  model: "claude-opus-4-6",

  url: "https://api.anthropic.com/v1/messages",

  provider: :anthropic,

  api_key_env: "ANTHROPIC_API_KEY",

  vision: true,

  receive_timeout: 120_000,

  # Step 10: Anthropic's `thinking` block `budget_tokens` field. This is

  # the only Anthropic-specific resource constraint. nil = Anthropic API

  # rejects the request. No Eai-side fallback (we don't add code for

  # provider compatibility). Override per-model or per-call via

  # `Eai.Chat.talk(reasoning_budget_tokens: N)` is NOT supported in this

  # step; the field is model-config only.

  reasoning_budget_tokens: 5000,

  # Step 10: `max_tokens` is also pass-through (no `|| 8192` fallback in

  # the adapter). Anthropic's Messages API requires this field — nil

  # would be rejected. The model config is the right place to declare

  # the resource limit. Override per-call via

  # `Eai.Chat.talk(max_tokens: N)` is plumbed through Step 7.

  max_tokens: 8192,

  # Step 9: opt into the `output-128k-2025-02-19` beta header so Opus 4.6+

  # can use 131072 max_tokens (without this header, Anthropic API caps

  # the request at 8192 for Opus 4.6 / Opus 4.7). Real Anthropic API

  # honours this header; DeepSeek `/anthropic` compatible endpoint drops

  # it silently. Override per-call via `Eai.Chat.talk(anthropic_beta: [...])`.

  anthropic_beta: ["output-128k-2025-02-19"]