Claude API

TAS Score: S0/3 — D2/5 — A0 / T0 (S0: Pause ❌ — service controlled by Anthropic. Exit ❌ — no model export. Recoverability ❌ — no rollback.) (D2: Personalisation ⚠️ conversation data processed. Hidden cost ⚠️ data retention policies. Transparency fragility ⚠️ proprietary model.)

Brief Description

Cloud-based AI API by Anthropic. Claude models (Opus, Sonnet, Haiku) accessible via API and web interface (claude.ai). Known for strong reasoning and safety alignment. No self-hosted option.

Why it’s in the catalog

Claude is one of the leading AI assistants. Including it makes the autonomy trade-off visible against self-hosted alternatives. This card exists for comparison, not recommendation.

Architectural Role

Cloud-hosted compute/inference. All processing on Anthropic/cloud infrastructure. API requires authentication and payment. Web interface requires account.

Technical Autonomy

  • Works without internet — no, cloud-only
  • Stores data locally — no, data on Anthropic servers
  • Does not require external accounts — requires Anthropic account
  • Allows data export — conversation export exists, model cannot be exported
  • Provides offline updates — no

Philosophical Assessment (whose.world criteria)

Criterion Status Comments
Pause ❌ Cannot pause the service. Anthropic controls availability.
Exit ❌ Conversation export possible. Model and capabilities cannot be taken.
Recoverability ❌ No rollback. Model versions change without user control.
Visibility ❌ Proprietary. No source code. No model weights.
External Dependencies ❌ Entirely dependent on Anthropic infrastructure.

Configuration (Minimal) 

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "content-type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{"model":"claude-sonnet-4-20250514","max_tokens":1024,"messages":[{"role":"user","content":"Hello"}]}'

No self-hosted option exists.

Autonomous Alternatives

Alternative Autonomy Notes
Ollama + Open WebUI A3 / T2 Self-hosted chat. Free, private, offline.
MiroThinker A3 / T2 Open-weight reasoning model.
DeepSeek-R1 A3 / T2 Open-weight reasoning. MIT licensed.

Trajectory

Direction: stable

Anthropic positions itself as a safety-focused AI lab. No open-weight models. API-only access. Pricing competitive but cloud-locked. Constitutional AI approach is documented but model weights are proprietary.

Signal assessment:

Signal Status Evidence
License ⚠️ Proprietary. No open-weight models.
Feature gating ⚠️ Free tier limited. Best models behind paid API.
Self-hosting ⚠️ No self-hosted option. API-only.
Governance ➖ Corporate. Safety-focused research published, but models closed.

Signal key: ✅ opening · ➖ neutral · ⚠️ closing

Sources

This site uses Just the Docs, a documentation theme for Jekyll.