Stream Anthropic responses uncompressed

[xymbol]

What this does

Net::HTTP requests gzip and auto-inflates by default. Cloudflare gzips Anthropic's SSE with infrequent deflate flushes, batching chunk delivery into 2 bursts and pushing first-chunk arrival to ~15s on a 22s response.

Setting Accept-Encoding: identity on streaming requests bypasses Net::HTTP's inflater. Scoped to Anthropic streaming; non-streaming responses still benefit from gzip.

Measured on claude-haiku-4-5, 1500-word completion. Sparkline: each char = 1s, digit = chunks delivered, _ = zero.

	TTFT	chunks/sec
before	15.1s	9____9
after	1.17s	84528454645464555__92

Type of change

• Bug fix
• New feature
• Breaking change
• Documentation
• Performance improvement

Scope check

• I read the Contributing Guide
• This aligns with RubyLLM's focus on LLM communication
• This isn't application-specific logic that belongs in user code
• This benefits most users, not just my specific use case

Required for new features

N/A — bug fix.

Quality check

• I ran overcommit --install and all hooks pass
• I tested my changes thoroughly
  • For provider changes: Re-recorded VCR cassettes with bundle exec rake vcr:record[provider_name] (not needed — VCR's default matcher is [:method, :uri], and Net::HTTP still inflates a recorded gzipped response on its own. The only load-bearing diff would be the Accept-Encoding request header value.)
  • All tests pass: bundle exec rspec
• I updated documentation if needed (no docs changes needed)
• I didn't modify auto-generated files manually (models.json, aliases.json)

New spec at spec/ruby_llm/providers/anthropic/streaming_spec.rb asserts the request header (fails without the fix, passes with).

AI-generated code

• I used AI tools to help write this code
• I have reviewed and understand all generated code (required if above is checked)

API changes

• Breaking change
• New public methods/classes
• Changed method signatures
• No API changes

Related

anthropics/anthropic-sdk-ruby#182 — the official Anthropic Ruby SDK has the same Net::HTTP auto-inflate bug and applies the same one-header fix (Accept-Encoding: identity) on its streaming endpoints.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.90%. Comparing base (80fe294) to head (c462d94).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #771      +/-   ##
==========================================
+ Coverage   82.88%   82.90%   +0.02%     
==========================================
  Files         142      142              
  Lines        6574     6576       +2     
  Branches     1148     1148              
==========================================
+ Hits         5449     5452       +3     
  Misses        663      663              
+ Partials      462      461       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[xymbol]

Thanks @crmne for your work on RubyLLM. 🙏

On #771 an Anthropic maintainer (@ms-jpq) confirmed the same bug and applied the same one-header fix in the official SDK (Accept-Encoding: identity). Rebased and green; glad to help if anything needs reshaping.