llm.rb 4.12.0 → 4.14.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (32) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +84 -0
  3. data/README.md +126 -736
  4. data/lib/llm/context.rb +12 -2
  5. data/lib/llm/error.rb +4 -0
  6. data/lib/llm/eventhandler.rb +16 -12
  7. data/lib/llm/eventstream/event.rb +15 -5
  8. data/lib/llm/eventstream/parser.rb +29 -14
  9. data/lib/llm/function.rb +1 -1
  10. data/lib/llm/mcp/command.rb +1 -1
  11. data/lib/llm/mcp/error.rb +31 -1
  12. data/lib/llm/mcp/mailbox.rb +23 -0
  13. data/lib/llm/mcp/pipe.rb +1 -1
  14. data/lib/llm/mcp/router.rb +44 -0
  15. data/lib/llm/mcp/rpc.rb +31 -15
  16. data/lib/llm/mcp/transport/http/event_handler.rb +11 -9
  17. data/lib/llm/mcp/transport/http.rb +2 -2
  18. data/lib/llm/mcp/transport/stdio.rb +1 -1
  19. data/lib/llm/mcp.rb +46 -2
  20. data/lib/llm/provider/transport/http/execution.rb +115 -0
  21. data/lib/llm/provider/transport/http/interruptible.rb +109 -0
  22. data/lib/llm/provider/transport/http/stream_decoder.rb +92 -0
  23. data/lib/llm/provider/transport/http.rb +144 -0
  24. data/lib/llm/provider.rb +17 -103
  25. data/lib/llm/providers/openai/request_adapter/respond.rb +11 -5
  26. data/lib/llm/providers/openai/response_adapter/responds.rb +13 -1
  27. data/lib/llm/providers/openai/responses/stream_parser.rb +31 -0
  28. data/lib/llm/version.rb +1 -1
  29. data/lib/llm.rb +8 -0
  30. data/llm.gemspec +16 -6
  31. metadata +23 -8
  32. data/lib/llm/client.rb +0 -36
checksums.yaml CHANGED
@@ -1,7 +1,7 @@
1
1
  ---
2
2
  SHA256:
3
- metadata.gz: 79d4a45ec25408e46451475575e917ef9d8579bec32f1a6a78bfed235e5ae212
4
- data.tar.gz: fdeb12175be3ef87e411021444305b9e785a9bf2d055dfdc7bf718f5740623d8
3
+ metadata.gz: ea1addf0bff644fa11e4f69a806f8ff5b7aa04fbbbc3f0592bd51b6ebc07f0f8
4
+ data.tar.gz: a3c846b9744e4ef230e2f23ed6ab42f6b4c84a0165b8bc066b7f6a003ee8fc00
5
5
  SHA512:
6
- metadata.gz: ea35b39b5476b75370485128dd8441e078bc7ac69236a7a50f4e32fb419f6fac5f7bb81faf3e029f28b788f4d69645e1b97e4126ea4f9fcc31f014921d2434a4
7
- data.tar.gz: c73bbf806f5cef71bfadfc1368fbdbfe07bf37118df18ebec71f4914a27ae2a3858fa6a210ee4d7cdff8f672a14c59016604a72a0a90c611b37223c4652ee991
6
+ metadata.gz: 7387da06d824d42753ff30455b0e464b7ca6eaa43e9410ce814ad96451c5595154d1e721fb69c9edc0971208aaf8a011ce42078827b57971e0e7c0a66eb0db6e
7
+ data.tar.gz: 590442f434086b7215d664e6b5d474130499a14fba16810ff7e0b04878d25e46ca8983057af5fd9275d8415d95da6e1439b84388fa450b8c06bc7841c832a48e
data/CHANGELOG.md CHANGED
@@ -1,5 +1,89 @@
1
1
  # Changelog
2
2
 
3
+ ## Unreleased
4
+
5
+ Changes since `v4.14.0`.
6
+
7
+ ## v4.14.0
8
+
9
+ Changes since `v4.13.0`.
10
+
11
+ This release adds request interruption for contexts, reworks provider
12
+ HTTP internals for lower-overhead streaming, and fixes MCP clients so
13
+ parallel tool calls can safely share one connection.
14
+
15
+ ### Add
16
+
17
+ * **Add request interruption support** <br>
18
+ Add `LLM::Context#interrupt!`, `LLM::Context#cancel!`, and
19
+ `LLM::Interrupt` for interrupting in-flight provider requests,
20
+ inspired by Go's context cancellation.
21
+
22
+ ### Change
23
+
24
+ * **Rework provider HTTP transport internals** <br>
25
+ Rework provider HTTP around `LLM::Provider::Transport::HTTP` with
26
+ explicit transient and persistent transport handling.
27
+
28
+ * **Reduce SSE parser overhead** <br>
29
+ Dispatch raw parsed values to registered visitors instead of building
30
+ an `Event` object for every streamed line.
31
+
32
+ * **Reduce provider streaming allocations** <br>
33
+ Decode streamed provider payloads directly in
34
+ `LLM::Provider::Transport::HTTP` before handing them to provider
35
+ parsers, which cuts allocation churn and gives a smaller streaming
36
+ speed bump.
37
+
38
+ * **Reduce generic SSE parser allocations** <br>
39
+ Keep unread event-stream buffer data in place until compaction is
40
+ worthwhile, which lowers allocation churn in the remaining generic
41
+ SSE path.
42
+
43
+ ### Fix
44
+
45
+ * **Support parallel MCP tool calls on one client** <br>
46
+ Route MCP responses by JSON-RPC id so concurrent tool calls can
47
+ share one client and transport without mismatching replies.
48
+
49
+ * **Use explicit MCP non-blocking read errors** <br>
50
+ Use `IO::EAGAINWaitReadable` while continuing to retry on
51
+ `IO::WaitReadable`.
52
+
53
+ ## v4.13.0
54
+
55
+ Changes since `v4.12.0`.
56
+
57
+ This release expands MCP prompt support, improves reasoning support in the
58
+ OpenAI Responses API, and refreshes the docs around llm.rb's runtime model,
59
+ contexts, and advanced workflows.
60
+
61
+ ### Add
62
+
63
+ - Add `LLM::MCP#prompts` and `LLM::MCP#find_prompt` for MCP prompt support.
64
+
65
+ ### Change
66
+
67
+ - Rework the README around llm.rb as a runtime for AI systems.
68
+ - Add a dedicated deep dive guide for providers, contexts, persistence,
69
+ tools, agents, MCP, tracing, multimodal prompts, and retrieval.
70
+
71
+ ### Fix
72
+
73
+ All of these fixes apply to MCP:
74
+
75
+ - fix(mcp): raise `LLM::MCP::MismatchError` on mismatched response ids.
76
+ - fix(mcp): normalize prompt message content while preserving the original payload.
77
+
78
+ All of these fixes apply to OpenAI's Responses API:
79
+
80
+ - fix(openai): emit `on_reasoning_content` for streamed reasoning summaries.
81
+ - fix(openai): skip `previous_response_id` on `store: false` follow-up calls.
82
+ - fix(openai): fall back to an empty object schema for tools without params.
83
+ - fix(openai): preserve original tool-call payloads on re-sent assistant tool messages.
84
+ - fix(openai): emit `output_text` for assistant-authored response content.
85
+ - fix(openai): return `nil` for `system_fingerprint` on normalized response objects.
86
+
3
87
  ## v4.12.0
4
88
 
5
89
  Changes since `v4.11.1`.