Skip to content
Home › Reference › LLM Providers › Anthropic

Anthropic

Configure Anthropic as the LLM provider for AMX's three sub-agents (Profile, RAG, Code). Claude is particularly strong at following structured-output instructions, which translates into cleaner column descriptions on schemas with awkward column names. This page walks through registering an Anthropic profile, picking a Claude model, choosing the right confidence thresholds for Claude's logprob-equivalent outputs, and confirming the profile is reachable.

Prerequisites

  • AMX installed (pip install amx-cli).
  • An Anthropic API key. Get one at console.anthropic.com.
  • A funded Anthropic account or enough free credit. AMX surfaces 429 / quota errors clearly but it cannot mint credits for you.
  • An active database profile (or follow Quick start first).

Step-by-step

1. Open the AMX REPL

amx

2. Add an LLM profile

> /add-llm-profile

Pick anthropic:

Select AI provider:
  openai
  openrouter
  anthropic
  ...
> anthropic

3. Answer the model + key prompts

Model: use the provider's natural model id. AMX will add any required provider prefix internally.
Anthropic model example: claude-sonnet-4-20250514
Model name: claude-sonnet-4-20250514
API key: ••••••••••••••••••••••••••••••••
Generation settings:
  Alternatives (1-5): 3
  Column batch size: 10
  Temperature (0.0-2.0): 0.2
Confidence thresholds (token probability 0.0-1.0):
  High threshold: 0.85
  Medium threshold: 0.50

Notes on each field:

  • Model name — type the bare Claude model id. AMX normalises internally.
  • API key — sk-ant-…. Stored in the OS keychain when one is available.
  • Alternatives / Batch size / Temperature — same defaults as OpenAI work fine; bump column_batch_size to 12–15 if you want to push Claude's longer-context strength.
  • Logprob thresholds — Claude doesn't return raw logprobs the way OpenAI does. AMX derives an equivalent from the model's stop-reason confidence and per-token sampling distribution. The default 0.85 / 0.50 is calibrated for claude-sonnet-4; relax for older Sonnet / Haiku models.

Which Claude model should I pick?

  • claude-sonnet-4-20250514 (default) — best quality / cost balance for AMX. Starts here.
  • claude-opus-4-… — highest quality for ambiguous schemas (legacy systems, transliterated column names). Slower and more expensive; use only after you've confirmed sonnet-4 isn't enough.
  • claude-haiku-3-5-… — lowest cost. Fine for whole-warehouse metadata-mode sweeps where you just need first-draft descriptions.
  • Older models (claude-3-5-sonnet, claude-3-opus) — supported but produce lower-quality descriptions on AMX's prompt template. Upgrade.

4. Activate and confirm

> /use-llm anthropic-prod
✓ Active LLM profile → anthropic-prod [anthropic] claude-sonnet-4-20250514

> /llm test
[anthropic] claude-sonnet-4-20250514 ... ✓ reached (latency: 821 ms, tokens: 14 in / 6 out)

5. (Optional) Enable extended thinking for hard schemas

For very ambiguous schemas (cryptic abbreviations, non-English column names), Claude's extended-thinking modes can substantially improve description quality at the cost of latency. Edit ~/.amx/config.yml and add thinking_budget_tokens under the profile:

llm_profiles:
  anthropic-deep:
    provider: anthropic
    model: claude-sonnet-4-20250514
    api_key: keyring://amx/anthropic-deep/api_key
    temperature: 0.2
    n_alternatives: 3
    column_batch_size: 8           # smaller batches when thinking is on
    thinking_budget_tokens: 4000   # thinking budget per request

Per-column cost roughly doubles, but on legacy schemas that's often the difference between a usable draft and one you'd rewrite from scratch.

6. Run a real description sweep

> /run sales.customer
[Profile] sampled scan on sales.customer ... ok (rows: 5000)
[LLM]     anthropic/claude-sonnet-4-20250514, batch 10, 18 columns ... ok in 5.8 s
          confidence: high 11 · medium 5 · low 2

Sample config

llm_profiles:
  anthropic-prod:
    provider: anthropic
    model: claude-sonnet-4-20250514
    api_key: keyring://amx/anthropic-prod/api_key
    temperature: 0.2
    n_alternatives: 3
    column_batch_size: 10
    logprob_high: 0.85
    logprob_medium: 0.50
active_llm_profile: anthropic-prod

Verify

  1. > /llm test — small ping completion. Surfaces auth / quota errors before a real /run.
  2. > /llm — confirms the active profile and model id.
  3. > amx doctor — confirms reachability and that the model id resolves.

Troubleshooting

Symptom Cause Fix
anthropic.AuthenticationError: invalid x-api-key Key revoked / typo Re-issue at console.anthropic.com; re-run /add-llm-profile
anthropic.RateLimitError: Number of requests exceeded … Per-minute request cap (Tier 1: 50 RPM) Lower column_batch_size so each request is a bigger payload, OR upgrade tier
anthropic.NotFoundError: model: claude-3-opus-20240229 not found Model id deprecated for your account Use claude-sonnet-4-20250514 (current default); confirm the available models in the Console
Lots of low confidence on otherwise simple columns Default thresholds calibrated for sonnet-4; older Sonnet / Haiku gives lower derived confidence /logprob-thresholds 0.7 0.4 for older models
anthropic.BadRequestError: max_tokens: … mid-/run Column batch + thinking budget exceeded the model's context window Lower column_batch_size or thinking_budget_tokens
Cost surprise Sonnet ≈ 3× Haiku per token; with n_alternatives: 3 you're paying for three drafts per column Drop to claude-haiku-3-5-… for sweeps, or use Batch mode for ~50% off

What's next

  • Batch mode — Anthropic's batch API for cheap async drafts (~50% off, 24h SLA).
  • OpenAI — same template; useful as a parallel profile for /history compare.
  • Run & Apply — review wizard keystrokes for picking between Claude's alternatives.