traceAI-livekit 0.1.3__tar.gz → 0.2.0__tar.gz

traceai_livekit-0.2.0/PKG-INFO

@@ -0,0 +1,221 @@
+Metadata-Version: 2.4
+Name: traceAI-livekit
+Version: 0.2.0
+Summary: OpenTelemetry instrumentation for LiveKit
+License: Apache-2.0
+Keywords: traceai,futureagi,opentelemetry,tracing,observability,llm,livekit
+Author: Future AGI
+Author-email: no-reply@futureagi.com
+Requires-Python: >=3.10,<3.14
+Classifier: Framework :: OpenTelemetry
+Classifier: Framework :: OpenTelemetry :: Instrumentations
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Classifier: Topic :: System :: Monitoring
+Requires-Dist: fi-instrumentation-otel (>=1.3.0)
+Project-URL: Changelog, https://github.com/future-agi/traceAI/blob/dev/python/frameworks/livekit/CHANGELOG.md
+Project-URL: Documentation, https://docs.futureagi.com/docs/integrations/traceai/livekit
+Project-URL: Homepage, https://futureagi.com
+Project-URL: Issues, https://github.com/future-agi/traceAI/issues
+Project-URL: Repository, https://github.com/future-agi/traceAI/tree/dev/python/frameworks/livekit
+Description-Content-Type: text/markdown
+
+# traceAI-livekit
+
+OpenTelemetry instrumentation for LiveKit agents integrated with Future AGI.
+
+This package provides automatic attribute mapping from LiveKit's native OpenTelemetry instrumentation to Future AGI's semantic conventions, ensuring rich and structured traces in your dashboard.
+
+## Installation
+
+```bash
+pip install traceai-livekit
+```
+
+## Quick Start
+
+To instrument your LiveKit agent, simply initialize `fi-instrumentation` and enable the attribute mapping **inside your agent's entrypoint**.
+
+```python
+from livekit.agents import AgentServer, JobContext
+from fi_instrumentation.otel import register, ProjectType
+from traceai_livekit import enable_http_attribute_mapping
+
+server = AgentServer()
+
+@server.rtc_session()
+async def entrypoint(ctx: JobContext):
+    # 1. Initialize TraceAI (inside entrypoint for multiprocessing safety)
+    register(
+        project_name="My LiveKit Agent",
+        project_type=ProjectType.OBSERVE,
+        set_global_tracer_provider=True,
+    )
+    
+    # 2. Enable LiveKit attribute mapping
+    # This automatically hooks into LiveKit's telemetry system
+    enable_http_attribute_mapping()
+    
+    # ... rest of your agent logic ...
+    await ctx.connect()
+
+if __name__ == "__main__":
+    server.run()
+```
+
+`enable_http_attribute_mapping()` (or `enable_grpc_attribute_mapping()`) hands the provider to LiveKit for you. If you call
+LiveKit's `set_tracer_provider` yourself on the provider `register()` returned, call it after `enable_*`: before it, the call
+drops the Future AGI exporter. LiveKit installs its privacy filter once per provider, so an `allow_pii` passed in a later call is
+ignored. To keep conversation content out of your traces, set `LIVEKIT_TELEMETRY_ALLOW_PII=false` before calling `enable_*`
+(livekit-agents 1.8 and later).
+
+### Flushing at the end of a job
+
+On Linux a LiveKit job process ends without running `atexit`, and LiveKit flushes only its own processors, so the spans
+still queued when a job ends, such as the root `agent_session` span, used to never reach Future AGI. `enable_*` now sends
+them twice over:
+
+- **When the job shuts down**, if `enable_*` runs inside a job (the usual place, the entrypoint): this sends the voice
+  trace, `agent_session` included, as soon as the call ends.
+- **When the process exits**, just before it stops: this sends what LiveKit ends after the shutdown callbacks, its own
+  `job_shutdown` span, and the job's root `job_entrypoint` span when tracing is set up before the job starts. With
+  set-up inside the entrypoint, LiveKit starts `job_entrypoint` before tracing is on, so it is never recorded and the
+  job trace arrives rooted at `job_shutdown`.
+
+Each waits at most 3 seconds for the collector, well inside the 10 seconds LiveKit gives a job process to exit, so its own
+clean-up still runs when the collector is slow.
+
+If you set tracing up outside a job (for example at import time, in the job process), only the process-exit flush runs.
+To also send each call's trace as soon as the call ends, call `flush_on_job_shutdown()` in each job entrypoint:
+
+```python
+from traceai_livekit import flush_on_job_shutdown
+
+@server.rtc_session()
+async def entrypoint(ctx: JobContext):
+    flush_on_job_shutdown()
+    ...
+```
+
+Do not pass `provider.force_flush` to `ctx.add_shutdown_callback`: LiveKit calls it with the shutdown reason, which
+`force_flush` takes as its timeout, so it fails and nothing is flushed.
+
+Limits:
+- A process LiveKit kills (for example when a worker drains and the job overruns its exit time) sends nothing more.
+- Before livekit-agents 1.3, LiveKit closes the session at the same time as the job-shutdown flush, so that flush can miss
+  `agent_session`; the process-exit flush still sends it.
+- A second `AgentSession` in one job that is still closing when the shutdown callbacks run is sent by the process-exit
+  flush, not the job-shutdown one.
+- In LiveKit Cloud observability mode the session's parent span is recorded on LiveKit's own provider and never reaches
+  Future AGI, so the trace has no stored root there.
+
+## Recording metadata on a session trace
+
+`record_egress` records metadata only: it never records, opens, downloads or uploads media. Each call creates and promptly ends one INTERNAL `livekit.recording.snapshot` span with FI kind `CHAIN`, parented to the supplied session span. Multiple recordings and start/stop/update responses remain separate snapshots instead of overwriting a root.
+
+Use an explicitly owned session root/envelope before starting recording. An ended span or a `NonRecordingSpan` holding its saved context is valid for later callbacks. Without `session_span`, the helper uses the current valid OTel span; it does not discover the native LiveKit root or search ancestors. Without a valid parent it returns `False` and creates no orphan recording trace.
+
+```python
+from fi_instrumentation import FITracer, TraceConfig
+from opentelemetry import trace
+from traceai_livekit import record_egress, wrap_egress_client
+
+# provider is the provider from register(); enable_* has already configured mapping.
+policy = TraceConfig()
+tracer = provider.get_tracer("my-agent-recordings")
+root = FITracer(tracer, config=policy).start_span(
+    "LiveKit Agent Session", fi_span_kind="agent"
+)
+saved_parent = trace.NonRecordingSpan(root.get_span_context())
+
+# Bind the egress SERVICE, not the owning API context manager. Requests and
+# response objects are forwarded unchanged; only successful responses are read.
+egress = wrap_egress_client(
+    api_owner.egress, session_span=saved_parent, room_name=room_name,
+    conversation_id=conversation_id,  # optional: may differ from room_name
+    tracer=tracer, trace_config=policy,
+)
+with trace.use_span(root, end_on_exit=False):
+    info = await egress.start_room_composite_egress(start_request)
+    # Start AgentSession under this context too, where the LiveKit mode supports it.
+
+# Later, for a final EgressInfo-like response ALREADY received by your application:
+recorded_locally = record_egress(
+    final_info, session_span=saved_parent, room_name=room_name,
+    conversation_id=conversation_id, tracer=tracer, trace_config=policy,
+)
+# End root at actual session shutdown; the application owns API stop/close.
+# See examples/egress_agent.py for a bounded, idempotent shutdown callback.
+```
+
+`True` means a snapshot was locally created and ended. It does **not** confirm collector delivery, recording completion, an available final file, or a playable URL. `False` also covers tracing suppression, unsampled children, invalid/missing ids, room mismatch and instrumentation failure. Delegate failures/cancellation are not swallowed or logged by the proxy. A failed or cancelled start with no returned id can leave server-side state unknown; the adapter does not retry or discover that state.
+
+### Metadata contract
+
+These are package-local custom attributes, not standardized recording semantics or a dashboard-player API:
+
+| FI export-copy attribute | Meaning |
+| --- | --- |
+| `livekit.recording.operation` | `snapshot` or a fixed captured method name |
+| `livekit.recording.egress_id` | Bounded nonempty response id |
+| `livekit.recording.status` | Recognized `EGRESS_*` label or decimal enum number; zero is retained, numbers are not interpreted |
+| `livekit.recording.started_at_unix_nano`, `livekit.recording.ended_at_unix_nano` | Positive int64 Unix nanoseconds, only when supplied |
+| `livekit.recording.room_name` | Explicit room, otherwise response room |
+| `gen_ai.conversation.id` | Explicit conversation id, independently masked; otherwise the fully masked room passed through conversation masks. A withheld room supplies no implicit conversation; a redacted room supplies only its replacement |
+| `livekit.recording.file_locations` | At most eight credential-free object locators or safe relative filenames |
+| `livekit.recording.stream_locations` | At most eight stream origins; paths may be bearer keys and are omitted |
+
+The helper's EgressInfo-like contract reads only `egress_id`, `room_name`, `status`, `started_at`, `ended_at`, `file_results` (`location`, or `filename` only when location is absent) and `stream_results` (`url`). Use a caller-normalized response if your SDK differs. Timestamps must already be Unix nanoseconds; there is no magnitude-based conversion. Missing/zero/invalid timestamps, unknown status text, unsupported legacy singular/segment/image results and absent locations are omitted. A successful start/stop may still be STARTING/ENDING with no final location; no duration or completion is inferred.
+
+Ids/room/conversation are at most 256 characters; locations at most 2048. Oversize values are omitted, not truncated. Only the first eight elements of each results collection are examined. A nonempty explicit room that differs from the response room refuses capture. No request, upload configuration, auth header, protobuf serialization or media bytes are inspected.
+
+### Privacy and adapter limits
+
+The recording helper controls only the metadata it writes. Attributes, events or links that your own samplers, span processors or code add are exported as written; the recording policy does not filter them, except the helper's reserved keys, which export always re-checks.
+
+- Helper-authored values pass existing FI `TraceConfig.mask` gates for both the destination alias and source key **before** span creation. Locations also pass the `output.value` policy, so output hiding applies. Both the helper policy and any supplied FITracer policy apply. A withheld helper value is not placed into attributes, events, links, exception text, logs, baggage, context keys, resources or the helper's boolean result.
+- FI export re-applies those policies by **key name**, not writer provenance, to the nine source/alias pairs in the metadata table (including `lk.pii.room_name` and `gen_ai.conversation.id`) and `session.id`. This also covers reserved attribute keys in application-created events and links; unrelated keys, event names, exception text and link targets are not filtered by the recording policy. Missing sources do not recreate aliases. A masking `Exception` replaces the affected value with `__REDACTED__` rather than exporting raw metadata or exception text. Export rechecking may transform an already-masked value again: custom masks should be idempotent.
+- Capture uses a per-call copy of the ordinary OTel SDK tracer without changing the supplied tracer/provider, sampler, or application callbacks. A lifecycle adapter retains the started child for `finally` cleanup and carries the policy tuple on a local instrumentation-scope copy for delayed/batched FI export. Normal scope fields and attributes are unchanged; no response values, policy data or identity registry are serialized. Application samplers/processors are not sandboxed or restricted. Unknown/custom tracers, subclasses, nested tracer proxies or inaccessible FI configuration fail closed before starting a span. Minimum-FI/SDK and LiveKit-copy compatibility still require verification.
+- Through creation **and ending**, the helper attaches a fresh context containing only the selected parent, including its trace state. It does not inherit ambient FI session/user/metadata attributes or baggage. Suppression is checked first; child termination and restoration of the caller context run in `finally`, including when `start_span` fails after SDK allocation. Initial extraction/masking failures still create no child.
+- The helper writes sensitive live-span values only under `lk.pii.*`; its aliases are created on the FI export copy. LiveKit PII opt-out and Cloud redaction can therefore remove helper-authored ids, room, conversation and locations before aliasing. The exact `livekit.recording.snapshot` and supported `livekit.recording.<method>` names remain reserved across tracer scopes: loss of the operation attribute to sampling or SDK limits cannot enable generic room-to-conversation fallback. Ordinary non-recording spans keep their existing mapping.
+- File locations allow `http`, `https`, `s3`, `gs` or safe relative names. Stream locations allow `http`, `https`, `rtmp`, `rtmps`, `rtsp`, `rtsps`, but only their origin survives. Malformed/unsupported locations, controls, backslashes and unsafe relative paths are omitted. All query parameters, fragments and userinfo are dropped. Credential-free locators are intentionally **not usable presigned playback links**. This stricter local projection supplements FI masks; it does not claim reuse of an independently verified shared URL/credential sanitizer.
+- The opt-in proxy recognizes coroutine functions named `start_room_composite_egress`, `start_web_egress`, `start_participant_egress`, `start_track_composite_egress`, `start_track_egress`, `stop_egress`, `update_stream`, and `update_layout`. This is a structural service contract, not a verified SDK-version matrix. Non-coroutine/decorated methods and unrelated attributes pass through; non-info responses produce no snapshot. Use the manual helper when recognition is unsupported.
+- The wrapper freezes its parent at creation; an unbound wrapper stays capture-disabled even if a different task later has a current span. Rewrapping with omitted/identical options is idempotent; conflicting supplied bindings raise a value-free `ValueError`. No global room registry or response history is kept.
+- Calls on the original service, other processes and server-side completion events are not intercepted. There is no polling or webhook server. Your application supplies final updates and distributed trace-context propagation. The API owner remains responsible for shutdown; special-method/context-manager protocols are not proxied.
+- The envelope is not a promise that FI receives LiveKit Cloud's native root. The earlier Cloud-root limitation still applies. Native `AgentSession` parent inheritance and real SDK compatibility must be validated for the deployed mode.
+
+The pinned development lock contains LiveKit Agents 1.8.6, API 1.2.1, protocol 1.1.27, FI 1.3.0 and OpenTelemetry SDK 1.44.0. Those pins are **verification targets, not executed compatibility results for this feature**. Fake-service tests do not establish SDK schemas, timestamp semantics, native-session linkage or a playback UI.
+
+## Compatibility
+
+*   **livekit-agents**: 1.x. 1.8 is the tested version (it needs Python 3.10 or later). Since 1.7, LiveKit names its content
+    attributes `lk.pii.*` (for example `lk.pii.user_input`); both those names and the earlier ones are mapped to `input.value`
+    and `output.value`.
+*   **Python**: 3.10 to 3.13.
+*   **fi-instrumentation-otel**: 1.3.0 or later; the recording helpers' privacy tests fail with 1.1.0. 1.3.0 needs Python 3.10
+    or later and sets the default base64 image cap to 0 (see its changelog). 1.0 moved span attributes from the `llm.*` names to the OpenTelemetry GenAI `gen_ai.*` names, and
+    renamed the `SpanAttributes` constants with them. Code that only calls `register()` is unaffected; if your code sets or reads
+    those attributes directly, see the
+    [mapping from the old names](https://github.com/future-agi/traceAI/blob/main/docs/OTEL_GENAI_SEMANTIC_CONVENTIONS.md#migration-from-openinference-llm).
+
+## Features
+
+*   **Automatic Attribute Mapping**: Converts LiveKit attributes (like `lk.user_input`, `gen_ai.usage.input_tokens`) to Future AGI standard attributes (`input.value`, `llm.tokenCount.prompt`).
+*   **Raw Data Preservation**: Keeps original LiveKit attributes and adds `raw.input` / `raw.output` for full debugging context.
+*   **Multiprocessing Support**: Designed to work with LiveKit's worker process model.
+*   **Event Parsing**: Extracts LLM outputs from `gen_ai.choice` events when available.
+
+## Examples
+
+Check the `examples/` directory for agent examples (validate recording SDK compatibility before deployment):
+*   `simple_agent.py`: A basic voice assistant agent.
+*   `egress_agent.py`: An agent with opt-in Egress metadata capture, an explicit session envelope and bounded stop/close cleanup. The modern `LiveKitAPI().egress` request shape and native-session linkage require the deployment's SDK/mode validation.
+

traceai_livekit-0.2.0/README.md

@@ -0,0 +1,190 @@
+# traceAI-livekit
+
+OpenTelemetry instrumentation for LiveKit agents integrated with Future AGI.
+
+This package provides automatic attribute mapping from LiveKit's native OpenTelemetry instrumentation to Future AGI's semantic conventions, ensuring rich and structured traces in your dashboard.
+
+## Installation
+
+```bash
+pip install traceai-livekit
+```
+
+## Quick Start
+
+To instrument your LiveKit agent, simply initialize `fi-instrumentation` and enable the attribute mapping **inside your agent's entrypoint**.
+
+```python
+from livekit.agents import AgentServer, JobContext
+from fi_instrumentation.otel import register, ProjectType
+from traceai_livekit import enable_http_attribute_mapping
+
+server = AgentServer()
+
+@server.rtc_session()
+async def entrypoint(ctx: JobContext):
+    # 1. Initialize TraceAI (inside entrypoint for multiprocessing safety)
+    register(
+        project_name="My LiveKit Agent",
+        project_type=ProjectType.OBSERVE,
+        set_global_tracer_provider=True,
+    )
+    
+    # 2. Enable LiveKit attribute mapping
+    # This automatically hooks into LiveKit's telemetry system
+    enable_http_attribute_mapping()
+    
+    # ... rest of your agent logic ...
+    await ctx.connect()
+
+if __name__ == "__main__":
+    server.run()
+```
+
+`enable_http_attribute_mapping()` (or `enable_grpc_attribute_mapping()`) hands the provider to LiveKit for you. If you call
+LiveKit's `set_tracer_provider` yourself on the provider `register()` returned, call it after `enable_*`: before it, the call
+drops the Future AGI exporter. LiveKit installs its privacy filter once per provider, so an `allow_pii` passed in a later call is
+ignored. To keep conversation content out of your traces, set `LIVEKIT_TELEMETRY_ALLOW_PII=false` before calling `enable_*`
+(livekit-agents 1.8 and later).
+
+### Flushing at the end of a job
+
+On Linux a LiveKit job process ends without running `atexit`, and LiveKit flushes only its own processors, so the spans
+still queued when a job ends, such as the root `agent_session` span, used to never reach Future AGI. `enable_*` now sends
+them twice over:
+
+- **When the job shuts down**, if `enable_*` runs inside a job (the usual place, the entrypoint): this sends the voice
+  trace, `agent_session` included, as soon as the call ends.
+- **When the process exits**, just before it stops: this sends what LiveKit ends after the shutdown callbacks, its own
+  `job_shutdown` span, and the job's root `job_entrypoint` span when tracing is set up before the job starts. With
+  set-up inside the entrypoint, LiveKit starts `job_entrypoint` before tracing is on, so it is never recorded and the
+  job trace arrives rooted at `job_shutdown`.
+
+Each waits at most 3 seconds for the collector, well inside the 10 seconds LiveKit gives a job process to exit, so its own
+clean-up still runs when the collector is slow.
+
+If you set tracing up outside a job (for example at import time, in the job process), only the process-exit flush runs.
+To also send each call's trace as soon as the call ends, call `flush_on_job_shutdown()` in each job entrypoint:
+
+```python
+from traceai_livekit import flush_on_job_shutdown
+
+@server.rtc_session()
+async def entrypoint(ctx: JobContext):
+    flush_on_job_shutdown()
+    ...
+```
+
+Do not pass `provider.force_flush` to `ctx.add_shutdown_callback`: LiveKit calls it with the shutdown reason, which
+`force_flush` takes as its timeout, so it fails and nothing is flushed.
+
+Limits:
+- A process LiveKit kills (for example when a worker drains and the job overruns its exit time) sends nothing more.
+- Before livekit-agents 1.3, LiveKit closes the session at the same time as the job-shutdown flush, so that flush can miss
+  `agent_session`; the process-exit flush still sends it.
+- A second `AgentSession` in one job that is still closing when the shutdown callbacks run is sent by the process-exit
+  flush, not the job-shutdown one.
+- In LiveKit Cloud observability mode the session's parent span is recorded on LiveKit's own provider and never reaches
+  Future AGI, so the trace has no stored root there.
+
+## Recording metadata on a session trace
+
+`record_egress` records metadata only: it never records, opens, downloads or uploads media. Each call creates and promptly ends one INTERNAL `livekit.recording.snapshot` span with FI kind `CHAIN`, parented to the supplied session span. Multiple recordings and start/stop/update responses remain separate snapshots instead of overwriting a root.
+
+Use an explicitly owned session root/envelope before starting recording. An ended span or a `NonRecordingSpan` holding its saved context is valid for later callbacks. Without `session_span`, the helper uses the current valid OTel span; it does not discover the native LiveKit root or search ancestors. Without a valid parent it returns `False` and creates no orphan recording trace.
+
+```python
+from fi_instrumentation import FITracer, TraceConfig
+from opentelemetry import trace
+from traceai_livekit import record_egress, wrap_egress_client
+
+# provider is the provider from register(); enable_* has already configured mapping.
+policy = TraceConfig()
+tracer = provider.get_tracer("my-agent-recordings")
+root = FITracer(tracer, config=policy).start_span(
+    "LiveKit Agent Session", fi_span_kind="agent"
+)
+saved_parent = trace.NonRecordingSpan(root.get_span_context())
+
+# Bind the egress SERVICE, not the owning API context manager. Requests and
+# response objects are forwarded unchanged; only successful responses are read.
+egress = wrap_egress_client(
+    api_owner.egress, session_span=saved_parent, room_name=room_name,
+    conversation_id=conversation_id,  # optional: may differ from room_name
+    tracer=tracer, trace_config=policy,
+)
+with trace.use_span(root, end_on_exit=False):
+    info = await egress.start_room_composite_egress(start_request)
+    # Start AgentSession under this context too, where the LiveKit mode supports it.
+
+# Later, for a final EgressInfo-like response ALREADY received by your application:
+recorded_locally = record_egress(
+    final_info, session_span=saved_parent, room_name=room_name,
+    conversation_id=conversation_id, tracer=tracer, trace_config=policy,
+)
+# End root at actual session shutdown; the application owns API stop/close.
+# See examples/egress_agent.py for a bounded, idempotent shutdown callback.
+```
+
+`True` means a snapshot was locally created and ended. It does **not** confirm collector delivery, recording completion, an available final file, or a playable URL. `False` also covers tracing suppression, unsampled children, invalid/missing ids, room mismatch and instrumentation failure. Delegate failures/cancellation are not swallowed or logged by the proxy. A failed or cancelled start with no returned id can leave server-side state unknown; the adapter does not retry or discover that state.
+
+### Metadata contract
+
+These are package-local custom attributes, not standardized recording semantics or a dashboard-player API:
+
+| FI export-copy attribute | Meaning |
+| --- | --- |
+| `livekit.recording.operation` | `snapshot` or a fixed captured method name |
+| `livekit.recording.egress_id` | Bounded nonempty response id |
+| `livekit.recording.status` | Recognized `EGRESS_*` label or decimal enum number; zero is retained, numbers are not interpreted |
+| `livekit.recording.started_at_unix_nano`, `livekit.recording.ended_at_unix_nano` | Positive int64 Unix nanoseconds, only when supplied |
+| `livekit.recording.room_name` | Explicit room, otherwise response room |
+| `gen_ai.conversation.id` | Explicit conversation id, independently masked; otherwise the fully masked room passed through conversation masks. A withheld room supplies no implicit conversation; a redacted room supplies only its replacement |
+| `livekit.recording.file_locations` | At most eight credential-free object locators or safe relative filenames |
+| `livekit.recording.stream_locations` | At most eight stream origins; paths may be bearer keys and are omitted |
+
+The helper's EgressInfo-like contract reads only `egress_id`, `room_name`, `status`, `started_at`, `ended_at`, `file_results` (`location`, or `filename` only when location is absent) and `stream_results` (`url`). Use a caller-normalized response if your SDK differs. Timestamps must already be Unix nanoseconds; there is no magnitude-based conversion. Missing/zero/invalid timestamps, unknown status text, unsupported legacy singular/segment/image results and absent locations are omitted. A successful start/stop may still be STARTING/ENDING with no final location; no duration or completion is inferred.
+
+Ids/room/conversation are at most 256 characters; locations at most 2048. Oversize values are omitted, not truncated. Only the first eight elements of each results collection are examined. A nonempty explicit room that differs from the response room refuses capture. No request, upload configuration, auth header, protobuf serialization or media bytes are inspected.
+
+### Privacy and adapter limits
+
+The recording helper controls only the metadata it writes. Attributes, events or links that your own samplers, span processors or code add are exported as written; the recording policy does not filter them, except the helper's reserved keys, which export always re-checks.
+
+- Helper-authored values pass existing FI `TraceConfig.mask` gates for both the destination alias and source key **before** span creation. Locations also pass the `output.value` policy, so output hiding applies. Both the helper policy and any supplied FITracer policy apply. A withheld helper value is not placed into attributes, events, links, exception text, logs, baggage, context keys, resources or the helper's boolean result.
+- FI export re-applies those policies by **key name**, not writer provenance, to the nine source/alias pairs in the metadata table (including `lk.pii.room_name` and `gen_ai.conversation.id`) and `session.id`. This also covers reserved attribute keys in application-created events and links; unrelated keys, event names, exception text and link targets are not filtered by the recording policy. Missing sources do not recreate aliases. A masking `Exception` replaces the affected value with `__REDACTED__` rather than exporting raw metadata or exception text. Export rechecking may transform an already-masked value again: custom masks should be idempotent.
+- Capture uses a per-call copy of the ordinary OTel SDK tracer without changing the supplied tracer/provider, sampler, or application callbacks. A lifecycle adapter retains the started child for `finally` cleanup and carries the policy tuple on a local instrumentation-scope copy for delayed/batched FI export. Normal scope fields and attributes are unchanged; no response values, policy data or identity registry are serialized. Application samplers/processors are not sandboxed or restricted. Unknown/custom tracers, subclasses, nested tracer proxies or inaccessible FI configuration fail closed before starting a span. Minimum-FI/SDK and LiveKit-copy compatibility still require verification.
+- Through creation **and ending**, the helper attaches a fresh context containing only the selected parent, including its trace state. It does not inherit ambient FI session/user/metadata attributes or baggage. Suppression is checked first; child termination and restoration of the caller context run in `finally`, including when `start_span` fails after SDK allocation. Initial extraction/masking failures still create no child.
+- The helper writes sensitive live-span values only under `lk.pii.*`; its aliases are created on the FI export copy. LiveKit PII opt-out and Cloud redaction can therefore remove helper-authored ids, room, conversation and locations before aliasing. The exact `livekit.recording.snapshot` and supported `livekit.recording.<method>` names remain reserved across tracer scopes: loss of the operation attribute to sampling or SDK limits cannot enable generic room-to-conversation fallback. Ordinary non-recording spans keep their existing mapping.
+- File locations allow `http`, `https`, `s3`, `gs` or safe relative names. Stream locations allow `http`, `https`, `rtmp`, `rtmps`, `rtsp`, `rtsps`, but only their origin survives. Malformed/unsupported locations, controls, backslashes and unsafe relative paths are omitted. All query parameters, fragments and userinfo are dropped. Credential-free locators are intentionally **not usable presigned playback links**. This stricter local projection supplements FI masks; it does not claim reuse of an independently verified shared URL/credential sanitizer.
+- The opt-in proxy recognizes coroutine functions named `start_room_composite_egress`, `start_web_egress`, `start_participant_egress`, `start_track_composite_egress`, `start_track_egress`, `stop_egress`, `update_stream`, and `update_layout`. This is a structural service contract, not a verified SDK-version matrix. Non-coroutine/decorated methods and unrelated attributes pass through; non-info responses produce no snapshot. Use the manual helper when recognition is unsupported.
+- The wrapper freezes its parent at creation; an unbound wrapper stays capture-disabled even if a different task later has a current span. Rewrapping with omitted/identical options is idempotent; conflicting supplied bindings raise a value-free `ValueError`. No global room registry or response history is kept.
+- Calls on the original service, other processes and server-side completion events are not intercepted. There is no polling or webhook server. Your application supplies final updates and distributed trace-context propagation. The API owner remains responsible for shutdown; special-method/context-manager protocols are not proxied.
+- The envelope is not a promise that FI receives LiveKit Cloud's native root. The earlier Cloud-root limitation still applies. Native `AgentSession` parent inheritance and real SDK compatibility must be validated for the deployed mode.
+
+The pinned development lock contains LiveKit Agents 1.8.6, API 1.2.1, protocol 1.1.27, FI 1.3.0 and OpenTelemetry SDK 1.44.0. Those pins are **verification targets, not executed compatibility results for this feature**. Fake-service tests do not establish SDK schemas, timestamp semantics, native-session linkage or a playback UI.
+
+## Compatibility
+
+*   **livekit-agents**: 1.x. 1.8 is the tested version (it needs Python 3.10 or later). Since 1.7, LiveKit names its content
+    attributes `lk.pii.*` (for example `lk.pii.user_input`); both those names and the earlier ones are mapped to `input.value`
+    and `output.value`.
+*   **Python**: 3.10 to 3.13.
+*   **fi-instrumentation-otel**: 1.3.0 or later; the recording helpers' privacy tests fail with 1.1.0. 1.3.0 needs Python 3.10
+    or later and sets the default base64 image cap to 0 (see its changelog). 1.0 moved span attributes from the `llm.*` names to the OpenTelemetry GenAI `gen_ai.*` names, and
+    renamed the `SpanAttributes` constants with them. Code that only calls `register()` is unaffected; if your code sets or reads
+    those attributes directly, see the
+    [mapping from the old names](https://github.com/future-agi/traceAI/blob/main/docs/OTEL_GENAI_SEMANTIC_CONVENTIONS.md#migration-from-openinference-llm).
+
+## Features
+
+*   **Automatic Attribute Mapping**: Converts LiveKit attributes (like `lk.user_input`, `gen_ai.usage.input_tokens`) to Future AGI standard attributes (`input.value`, `llm.tokenCount.prompt`).
+*   **Raw Data Preservation**: Keeps original LiveKit attributes and adds `raw.input` / `raw.output` for full debugging context.
+*   **Multiprocessing Support**: Designed to work with LiveKit's worker process model.
+*   **Event Parsing**: Extracts LLM outputs from `gen_ai.choice` events when available.
+
+## Examples
+
+Check the `examples/` directory for agent examples (validate recording SDK compatibility before deployment):
+*   `simple_agent.py`: A basic voice assistant agent.
+*   `egress_agent.py`: An agent with opt-in Egress metadata capture, an explicit session envelope and bounded stop/close cleanup. The modern `LiveKitAPI().egress` request shape and native-session linkage require the deployment's SDK/mode validation.

{traceai_livekit-0.1.3 → traceai_livekit-0.2.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "traceAI-livekit"
-version = "0.1.3"
+version = "0.2.0"
 description = "OpenTelemetry instrumentation for LiveKit"
 authors = ["Future AGI <no-reply@futureagi.com>"]
 readme = "README.md"
@@ -15,7 +15,6 @@ classifiers = [
     "Programming Language :: Python",
     "Programming Language :: Python :: 3",
     "Programming Language :: Python :: 3 :: Only",
-    "Programming Language :: Python :: 3.9",
     "Programming Language :: Python :: 3.10",
     "Programming Language :: Python :: 3.11",
     "Programming Language :: Python :: 3.12",
@@ -27,8 +26,8 @@ classifiers = [
 ]
 
 [tool.poetry.dependencies]
-python = ">3.9,<3.14"
-fi-instrumentation-otel = ">=1.1.0"
+python = ">=3.10,<3.14"
+fi-instrumentation-otel = ">=1.3.0"
 
 [tool.poetry.group.dev.dependencies]
 # livekit-agents 1.8 is the version whose privacy filter the tests cover; it needs Python 3.10+.

{traceai_livekit-0.1.3 → traceai_livekit-0.2.0}/traceai_livekit/__init__.py

@@ -19,7 +19,9 @@ from .integration import (
     flush_on_job_shutdown,
 )
 
-__version__ = "0.1.3"  # x-release-please-version
+from .recording import record_egress, wrap_egress_client
+
+__version__ = "0.2.0"  # x-release-please-version
 
 __all__ = [
     "enable_fi_attribute_mapping",
@@ -28,6 +30,8 @@ __all__ = [
     "create_mapped_http_exporter",
     "create_mapped_grpc_exporter",
     "flush_on_job_shutdown",
+    "record_egress",
+    "wrap_egress_client",
     # Exporters
     "MappedHTTPSpanExporter",
     "MappedGRPCSpanExporter",