add new features by shloimy-wiesel · Pull Request #41 · shloimy-wiesel/ai-sdk-stream-python

shloimy-wiesel · 2026-03-23T10:40:55Z

Fix tool name duplication in consume_openai_stream — only set buf["name"] when empty instead of appending
Fix tool call ID consistency — capture ToolCallHandle.toolCallId from begin_tool_call/start_tool_input and use it in ConsumeResult.tool_calls
Add tests for edge cases (repeated tool name, missing tool call ID) in both streaming and non-streaming modes
Run tests to validate changes (237/237 pass)
Code review and security scan (clean)

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.

Automatically generated by python-semantic-release

Adds an opt-in collect: bool = False parameter to StreamContext that accumulates all emitted content (text, reasoning, tool calls, sources, step count, finish reason) into a StreamRecord accessible via ctx.record. Useful for persisting conversation turns to a database after streaming.

feat: add collect=True to StreamContext for stream data recording

Automatically generated by python-semantic-release

* feat: add typed custom_information support to StreamContext Closes #6 - StreamContext is now Generic[_InfoT] (bound to pydantic.BaseModel) - New `custom_information` constructor parameter stores a read-only Pydantic model accessible via ctx.info throughout the stream lifecycle - Useful for carrying request-scoped metadata (user_id, rate_limit, etc.) through service layers without extra function arguments - 5 new tests; pyright passes with 0 errors * fix: correct docstring type and test organization for custom_information - Fix `info : _InfoT` → `info : _InfoT | None` in class Attributes docstring - Group custom_information tests into TestCustomInformation class

Automatically generated by python-semantic-release

Add ErrorEvent Pydantic model (type: "error", errorText: str), ctx.error(error_text) which emits the event then terminates the stream, and exports ErrorEvent from the top-level package. Resolves: #8

Add AbortEvent model (type: "abort", reason: str | None) and update ctx.abort() to accept an optional reason and emit the event before the [DONE] sentinel. Backward-compatible: bare ctx.abort() still works and omits the reason field from the wire output. Resolves: #13

Add FileEvent model (type: "file", url, mediaType), ctx.write_file() which auto-emits start/start-step before the event, FileRecord for collection, and files list on StreamRecord with to_dict() support. Resolves: #9

Add DataEvent model (dynamic data-{name} type), ctx.write_data() with name validation and transient flag, DataPartRecord for collection, and data_parts list on StreamRecord with to_dict() support. Transient parts are sent on the wire but not persisted to ctx.record. Resolves: #10

Add start_tool_input(), stream_tool_input_delta(), and finish_tool_input() to StreamContext. start_tool_input emits tool-input-start and returns a ToolCallHandle; each stream_tool_input_delta emits a tool-input-delta; finish_tool_input emits tool-input-available and updates the collected ToolCallRecord with the final input dict. begin_tool_call() is unchanged. Resolves: #11

Automatically generated by python-semantic-release

* Initial plan * feat: add on_finish callback for post-stream persistence Co-authored-by: shloimy-wiesel <144027408+shloimy-wiesel@users.noreply.github.com> Agent-Logs-Url: https://github.com/shloimy-wiesel/ai-sdk-stream-python/sessions/5f2f939a-91ee-46f4-9d8e-5abc6ef8a02e --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: shloimy-wiesel <144027408+shloimy-wiesel@users.noreply.github.com>

Automatically generated by python-semantic-release

…vent `messageMetadata` (#26) feat: support start_metadata on StreamContext for start event messageMetadata Co-authored-by: shloimy-wiesel <144027408+shloimy-wiesel@users.noreply.github.com> Agent-Logs-Url: https://github.com/shloimy-wiesel/ai-sdk-stream-python/sessions/e1fc71ca-6c44-48e5-9f74-df0da0cb80f5 --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: shloimy-wiesel <144027408+shloimy-wiesel@users.noreply.github.com>

When a background task raises an unhandled exception the stream now terminates with an error event instead of blocking the client forever. run() also auto-calls finish() if the coroutine returns without it. Resolves: #20

Adds a new optional contrib module that eliminates the ~40-line boilerplate every OpenAI + StreamContext backend must write. - Maps delta.content → ctx.write_text() - Maps delta.reasoning / delta.reasoning_content → ctx.write_reasoning() - Buffers tool call chunks by index; emits tool-input-start + tool-input-available at end (or streams deltas when stream_tool_deltas=True) - Extracts finish_reason and usage (when stream_options include_usage=True) - No hard dependency on the openai package — uses duck typing throughout - Returns ConsumeResult with content, tool_calls, finish_reason, usage Closes #23

…izer Closes #17 - Add `reasoning_tokens`, `answer_tokens`, `prompt_tokens` fields to `StreamRecord` - Add `total_output_tokens` and `total_tokens` computed properties - `write_text` and `write_reasoning` now auto-increment token counts via configurable `count_func` (defaults to `len`, i.e. character count) - Add `StreamContext(count_func=...)` parameter to swap in any `Callable[[str], int]` (e.g. tiktoken, word count) - Add `ctx.set_usage(prompt_tokens, reasoning_tokens, answer_tokens)` to override auto-counted values with exact LLM-reported numbers - Include all token fields in `StreamRecord.to_dict()`

feat: add timing (created_at/finished_at/duration_ms) to StreamRecord Co-authored-by: shloimy-wiesel <144027408+shloimy-wiesel@users.noreply.github.com> Agent-Logs-Url: https://github.com/shloimy-wiesel/ai-sdk-stream-python/sessions/781f63ee-0c28-4c82-bcc1-73f004039b4d * fix: address review comments on StreamRecord timing fields - Remove dead-code guard in to_dict(): created_at is non-optional so the `if self.created_at else None` branch was never reachable - Document created_at, finished_at, and duration_ms in StreamRecord docstring - Rename test_finished_at_none_before_finish → test_finished_at_none_initially to avoid confusion (the test called finish() internally) --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: shloimy-wiesel <144027408+shloimy-wiesel@users.noreply.github.com> Co-authored-by: shloimy wiesel <w.63071@gmail.com>

…cstring Expand the ctx.run() docstring to describe its three safety guarantees (auto-finish, auto-error, task GC prevention) with a full FastAPI example. Update all README examples to use ctx.run() instead of manual try/finally + asyncio.create_task(). Add a dedicated "ctx.run() — safe task runner" section to the StreamContext API docs. Closes #29

Add a `collect` keyword parameter (default True) to write_text, write_reasoning, write_source, write_file, begin_tool_call, and start_tool_input. When collect=False, the event is still emitted to the SSE stream but not recorded in ctx.record. This lets developers stream ephemeral content (status messages, internal tool calls) without polluting the persisted record. Closes #30

… context Change per-call collect parameter default from True to None (follow context setting). When a caller explicitly passes collect=True but the StreamContext was created with collect=False, raise RuntimeError instead of silently discarding the data. This prevents a class of silent bugs where developers think they are collecting but the context has no record. Three-way semantics: - None (default): collect if context-level collection is enabled - True: require collection; raise if no record exists - False: skip collection even if context-level is enabled

The AI SDK v6 wire protocol allows the `data` field in `data-{name}` events to be any JSON value (string, number, null, array, or object). The previous `dict[str, Any]` constraint made common patterns like streaming chat titles or signaling artifact completion impossible. Closes #34

…aEvent.encode() - write_data() now accepts collect: bool | None = None and routes through _should_collect(), consistent with all other high-level write helpers - DataEvent.encode() uses model_dump_json() (Pydantic's serializer) instead of model_dump() + json.dumps(), fixing TypeError for non-JSON-native types (datetime, UUID, Enum, etc.) inside the data field

…json-value fix: write_data() accepts any JSON-serializable value, not just dict

Automatically generated by python-semantic-release

vercel · 2026-03-23T10:40:59Z

Someone is attempting to deploy this pull request to the Shloimy's projects Team on Vercel.

No GitHub account was found matching the commit author email address.

To deploy this pull request, the commit author's email address needs to be associated with a GitHub account.

Learn more about how to change the commit author information.

Automatically generated by python-semantic-release

Converts AI SDK v6 UIMessage parts-based format to list[ChatCompletionMessageParam] for the OpenAI API, completing the input → output pipeline in contrib.openai alongside consume_openai_stream(). Part mapping: - TextUIPart → content string or {"type":"text"} block - FileUIPart (image) → {"type":"image_url"} content block - ToolUIPart (output-available) → assistant tool_calls + role:"tool" message - ToolUIPart (output-error) → assistant tool_calls + tool error message - ToolUIPart (other states) → assistant tool_calls only - ReasoningUIPart → dropped by default; included in <reasoning> tags when include_reasoning=True Closes #31.

Automatically generated by python-semantic-release

- Root AGENTS.md: comprehensive overview of what library solves, state machine architecture, event flow, wire protocol, capabilities, conventions, anti-patterns - src/ai_sdk_stream_python/AGENTS.md: internal implementation details, module map, method→event mapping, collection fields, OpenAI adapter internals

Copilot

Pull request overview

This PR expands ai-sdk-stream-python toward a fuller AI SDK v6 backend toolkit by adding request-body Pydantic types, stream data collection, and optional OpenAI integration helpers alongside new event types and richer StreamContext capabilities.

Changes:

Add StreamContext features: collection (collect=True + ctx.record), ctx.run() safe runner, custom info (ctx.info), streaming tool-input deltas, and error/abort/data/file helpers.
Add new protocol models/events (file, error, abort, dynamic data-*) and request deserialization models (ChatRequest, UIMessage, *UIPart).
Add contrib.openai adapter utilities with extensive tests and documentation updates.

Reviewed changes

Copilot reviewed 16 out of 17 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
`uv.lock`	Updates editable package version entry (currently mismatched with `pyproject.toml`).
`pyproject.toml`	Bumps project version and defines build/test/tooling configuration.
`README.md`	Documents `ctx.run()`, collection, and `ctx.info` usage and API guidance.
`CHANGELOG.md`	Records new alpha releases and feature additions.
`AGENTS.md`	Project-level knowledge base for architecture and conventions.
`src/ai_sdk_stream_python/AGENTS.md`	Library-internals documentation and state machine/event mapping details.
`src/ai_sdk_stream_python/context.py`	Core `StreamContext` enhancements: collection, callbacks, metadata, data/file/error/abort, tool delta streaming, `run()`.
`src/ai_sdk_stream_python/events.py`	Adds `FileEvent`, `ErrorEvent`, `AbortEvent`, and `DataEvent` encode behavior; updates union exports.
`src/ai_sdk_stream_python/collect.py`	Introduces `StreamRecord` and related record dataclasses for persistence/auditing.
`src/ai_sdk_stream_python/types.py`	Adds Pydantic models for incoming AI SDK v6 request payloads.
`src/ai_sdk_stream_python/contrib/openai.py`	Adds OpenAI stream consumer and UIMessage→OpenAI message conversion utilities.
`src/ai_sdk_stream_python/contrib/__init__.py`	Introduces contrib package marker/docstring.
`src/ai_sdk_stream_python/__init__.py`	Re-exports new public API surface (records, types, new events, callback type).
`tests/test_context.py`	Extends context tests for new tool delta streaming, data/file, abort/error, callbacks, metadata, and `run()`.
`tests/test_collect.py`	Adds comprehensive tests for `collect=True` and per-call collect tri-state semantics.
`tests/test_types.py`	Adds tests for request-body Pydantic models and discriminated unions.
`tests/test_contrib_openai.py`	Adds extensive tests for OpenAI adapter behavior (text/reasoning/tool calls/messages conversion).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-29T10:11:02Z

+
+            fn_name = getattr(fn, "name", None)
+            if fn_name:
+                buf["name"] += fn_name


fn.name is appended to buf["name"], which can duplicate the tool name if the stream repeats the name on multiple chunks (resulting in e.g. searchsearch). Consider only setting the name when it’s empty, or otherwise de-duplicating, while still supporting truly incremental name streaming if needed.

Suggested change

buf["name"] += fn_name

# Avoid duplicating the tool name across chunks.

# If we don't have a name yet, just set it.

if not buf["name"]:

buf["name"] = fn_name

else:

# Support incremental name streaming (e.g. "se" -> "sea" -> "search"):

# if the new chunk starts with the existing name, treat it as an update.

if fn_name.startswith(buf["name"]):

buf["name"] = fn_name

# If we already have the full name (or a longer variant), do nothing.

elif buf["name"].endswith(fn_name):

pass

else:

# Fallback: append only when it doesn't clearly duplicate.

buf["name"] += fn_name

Copilot · 2026-03-29T10:11:03Z

+            if stream_tool_deltas:
+                # Start streaming once the tool name is known
+                if buf["handle"] is None and buf["name"]:
+                    buf["handle"] = await ctx.start_tool_input(
+                        buf["name"],
+                        tool_call_id=buf["id"] or None,
+                    )
+                if buf["handle"] is not None:
+                    await ctx.stream_tool_input_delta(
+                        buf["handle"].toolCallId,
+                        fn_args,
+                    )
+
+    # ── Finalize tool calls ──────────────────────────────────────────────────
+    result_tool_calls: list[dict[str, str]] = []
+    for idx in sorted(tool_buffer):
+        buf = tool_buffer[idx]
+        try:
+            parsed_input: dict[str, Any] = json.loads(buf["arguments"] or "{}")
+        except json.JSONDecodeError:
+            parsed_input = {}
+
+        if stream_tool_deltas:
+            if buf["handle"] is not None:
+                await ctx.finish_tool_input(
+                    buf["handle"].toolCallId,
+                    buf["name"],
+                    parsed_input,
+                )
+            else:
+                # No arguments were streamed (zero-arg tool or empty chunks);
+                # fall back to the non-streaming path.
+                await ctx.begin_tool_call(
+                    buf["name"],
+                    parsed_input,
+                    tool_call_id=buf["id"] or None,
+                )
+        else:
+            await ctx.begin_tool_call(
+                buf["name"],
+                parsed_input,
+                tool_call_id=buf["id"] or None,
+            )
+
+        result_tool_calls.append(
+            {
+                "id": buf["id"],
+                "name": buf["name"],
+                "arguments": buf["arguments"],
+            }
+        )


When tc.id is missing (or arrives after start_tool_input is called), StreamContext will generate a new toolCallId but ConsumeResult.tool_calls still records id as an empty string / possibly a different value. Capture and reuse the actual ToolCallHandle.toolCallId returned by start_tool_input/begin_tool_call so emitted events and the returned ConsumeResult stay consistent.

shloimy-wiesel · 2026-03-29T10:24:34Z

@copilot apply changes based on the comments in this thread
if anything else needs changes/improvements fix them too

…y in consume_openai_stream Agent-Logs-Url: https://github.com/shloimy-wiesel/ai-sdk-stream-python/sessions/f0d42dd6-0fe4-44ed-be7e-1e5e9abaaf43 Co-authored-by: shloimy-wiesel <144027408+shloimy-wiesel@users.noreply.github.com>

Copilot · 2026-03-29T10:29:28Z

@copilot apply changes based on the comments in this thread
if anything else needs chang...

Both review comments have been addressed in 6351182:

Tool name duplication — Changed buf["name"] += fn_name to only set the name when the buffer is empty (if fn_name and not buf["name"]), preventing names like "searchsearch" when the stream repeats the name across chunks.
Tool call ID consistency — Now captures the ToolCallHandle.toolCallId returned by begin_tool_call()/start_tool_input() and uses it in ConsumeResult.tool_calls instead of buf["id"]. This ensures the result always matches the IDs emitted in SSE events, even when no tc.id is provided and StreamContext generates a UUID.

Added 4 new tests covering both edge cases in streaming and non-streaming modes. All 237 tests pass.

…xample only) - Implemented 5 tools: getWeather, createDocument, updateDocument, editDocument, requestSuggestions - Fixed conversation history persistence: assistant messages now saved to Redis - Uses StreamContext for lifecycle-safe streaming - Redis for session persistence (chat history + documents) - Handles delta strategy per artifact kind (text=incremental, code/sheet=full) - All tools verified working via browser Playwright tests

Automatically generated by python-semantic-release

semantic-release and others added 29 commits March 22, 2026 09:42

0.1.2-a.1

964c416

Automatically generated by python-semantic-release

docs: add stream collection feature to README.md

4fc9879

Merge pull request #4 from shloimy-wiesel/feat/stream-collect

75a5086

feat: add collect=True to StreamContext for stream data recording

0.2.0-a.1

3a8b972

Automatically generated by python-semantic-release

0.2.0-a.2

7b7a288

Automatically generated by python-semantic-release

feat: add error event type and ctx.error() helper

13836d4

Add ErrorEvent Pydantic model (type: "error", errorText: str), ctx.error(error_text) which emits the event then terminates the stream, and exports ErrorEvent from the top-level package. Resolves: #8

feat: add file event type and ctx.write_file() helper

d11d6ec

Add FileEvent model (type: "file", url, mediaType), ctx.write_file() which auto-emits start/start-step before the event, FileRecord for collection, and files list on StreamRecord with to_dict() support. Resolves: #9

0.2.0-a.3

552fd1c

Automatically generated by python-semantic-release

0.2.0-a.4

41d3949

Automatically generated by python-semantic-release

docs: add project guidelines and conventions to AGENTS.md

de33643

chore: format code

0b1f29a

Merge pull request #37 from shloimy-wiesel/fix/write-data-accept-any-…

954a077

…json-value fix: write_data() accepts any JSON-serializable value, not just dict

0.2.0-a.5

b20296c

Automatically generated by python-semantic-release

vercel Bot deployed to Preview March 23, 2026 10:45 View deployment

0.2.0-a.6

0bd070d

Automatically generated by python-semantic-release

vercel Bot had a problem deploying to Preview March 23, 2026 10:45 Failure

vercel Bot deployed to Preview March 23, 2026 10:57 View deployment

0.2.0-a.7

43b0ad2

Automatically generated by python-semantic-release

vercel Bot had a problem deploying to Preview March 23, 2026 10:58 Failure

vercel Bot deployed to Preview March 23, 2026 18:35 View deployment

shloimy-wiesel requested a review from Copilot March 29, 2026 10:06

Copilot started reviewing on behalf of shloimy-wiesel March 29, 2026 10:06 View session

Copilot AI reviewed Mar 29, 2026

View reviewed changes

Copilot started work on behalf of shloimy-wiesel March 29, 2026 10:24 View session

Copilot finished work on behalf of shloimy-wiesel March 29, 2026 10:30

vercel Bot deployed to Preview March 29, 2026 10:30 View deployment

Merge branch 'main' into alpha

c26c22b

vercel Bot deployed to Preview March 29, 2026 10:33 View deployment

shloimy-wiesel added 2 commits March 29, 2026 10:35

chore: run format

71138a9

vercel Bot deployed to Preview March 29, 2026 10:37 View deployment

chore: run format

2868db4

vercel Bot deployed to Preview March 29, 2026 10:39 View deployment

0.2.0-a.8

a0f7c75

Automatically generated by python-semantic-release

vercel Bot had a problem deploying to Preview March 29, 2026 10:39 Failure

shloimy-wiesel marked this pull request as ready for review March 29, 2026 10:42

shloimy-wiesel merged commit 749cab2 into main Mar 29, 2026
1 of 3 checks passed

shloimy-wiesel deleted the alpha branch March 29, 2026 10:43

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add new features#41

add new features#41
shloimy-wiesel merged 40 commits into
mainfrom
alpha

shloimy-wiesel commented Mar 23, 2026 •

edited by Copilot AI

Loading

Uh oh!

vercel Bot commented Mar 23, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 29, 2026

Uh oh!

Copilot AI Mar 29, 2026

Uh oh!

shloimy-wiesel commented Mar 29, 2026

Uh oh!

Copilot AI commented Mar 29, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

-                buf["name"] += fn_name
+                # Avoid duplicating the tool name across chunks.
+                # If we don't have a name yet, just set it.
+                if not buf["name"]:
+                    buf["name"] = fn_name
+                else:
+                    # Support incremental name streaming (e.g. "se" -> "sea" -> "search"):
+                    # if the new chunk starts with the existing name, treat it as an update.
+                    if fn_name.startswith(buf["name"]):
+                        buf["name"] = fn_name
+                    # If we already have the full name (or a longer variant), do nothing.
+                    elif buf["name"].endswith(fn_name):
+                        pass
+                    else:
+                        # Fallback: append only when it doesn't clearly duplicate.
+                        buf["name"] += fn_name

Conversation

shloimy-wiesel commented Mar 23, 2026 • edited by Copilot AI Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vercel Bot commented Mar 23, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 29, 2026

Choose a reason for hiding this comment

Uh oh!

shloimy-wiesel commented Mar 29, 2026

Uh oh!

Copilot AI commented Mar 29, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

shloimy-wiesel commented Mar 23, 2026 •

edited by Copilot AI

Loading