What pydantic/logfire shipped
Written by FoxPlug from the repository's own releases; not affiliated with Pydantic. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/pydantic/logfire; Pydantic did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Added a fleet option to logfire.configure so a Fleet policy can govern what the SDK sends. Pull request #2736
- Logfire.configure now sends telemetry with the API key when no write token is configured. Pull request #2743
- Agent_sandboxes now connects over gRPC and requires pydantic-monty-client 1.5.0 or later. Pull request #2737
- Agent_sandboxes now returns a plain AsyncMontyRemote instead of the AgentSandboxPool wrapper. Pull request #2756
- Agent_sandboxes now raises typed Agent Sandboxes errors so callers can tell a refused checkout from a connection failure. Pull request #2747
- The SDK now requires a project write token or saved project credentials before it sends data. Pull request #2683
- Logfire run now falls back to OpenTelemetry instrumentors for packages without an instrument method. Pull request #2609
- Streamed LLM responses stay iterable when recording them fails. Pull request #2721
- Dataset cases with dictionary-valued inputs, expected outputs and metadata are now serialized correctly. Pull request #2713
- The console exporter no longer crashes on attributes that format to an empty string when colors are off. Pull request #2731
Why it matters
This week the SDK gained a way to manage it from Fleet and a simpler credential path, so an Agent Sandboxes user needs only the API key. It also made agent sandboxes connect over gRPC with typed errors, and fixed several recording and export edge cases that users could hit in normal use.
Changelog entry
- Add a fleet option to logfire.configure so a Fleet policy governs what the SDK sends. Pull request #2736
- Send telemetry with the API key when no write token is configured. Pull request #2743
- Connect agent sandboxes over gRPC and require pydantic-monty-client 1.5.0. Pull request #2737
- Return a plain AsyncMontyRemote from logfire.agent_sandboxes(). Pull request #2756
- Raise typed Agent Sandboxes errors from logfire.agent_sandboxes(). Pull request #2747
- Require project credentials before the SDK sends data. Pull request #2683
- Fall back to OpenTelemetry instrumentors in logfire run. Pull request #2609
- Keep streamed LLM responses iterable when recording them fails. Pull request #2721
- Serialize dictionary-valued dataset case fields. Pull request #2713
- Stop the console exporter crashing on attributes that format to an empty string. Pull request #2731
- Make SpanLevel compare equal to warning. Pull request #2732
- Skip empty items in comma-separated Pydantic plugin include and exclude settings. Pull request #2720
Logfire v5.5.0 is out. logfire.configure takes fleet to let a policy govern what the SDK sends, the API key alone can carry telemetry, agent_sandboxes now connects over gRPC and raises typed errors, and logfire run falls back to OpenTelemetry instrumentors.
This week's release brings logfire.configure(fleet=...) so a Fleet policy governs sampling and what the SDK sends, and configure no longer needs a write token when an API key is set. agent_sandboxes now connects over gRPC, returns a plain AsyncMontyRemote, and raises typed errors so refused checkouts can be told apart from connection failures. logfire run now falls back to OpenTelemetry instrumentors for packages without an instrument method, and session replay, dataset serialization and console output edge cases are fixed.
Week of October 26, 2026
What shipped
- Version 5.4.0 was released, adding logfire.agent_sandboxes() and several fixes across path serialization, the console exporter and the secret scrubber. Release
- AWS security tokens, including X-Amz-Security-Token and
aws_security_token, are now scrubbed by default. Pull request #2692 - Thread pool tasks now carry managed variable overrides and the targeting context along with trace and baggage. Pull request #2680
- Setting
distributed_tracingto false now drops all incoming context extraction, including baggage and values from custom propagators. Pull request #2679 - Errors raised while consuming an instrumented LLM stream are now recorded on the streaming response log with exception details. Pull request #2491
- The read-tokens create command with save no longer crashes on Windows, where os.fchmod does not exist. Pull request #2696
- A new
console_logoption avoids duplicate console output when Loguru is used alongside Logfire. Pull request #2690 - Remote strings in log messages are hardened by escaping non-printable characters and treating MCP server logger names as fixed templates. Pull request #2654
- Binary attribute values such as bytes and bytearray are no longer rewritten into lists of integers by the secret scrubber. Pull request #2668
- All pathlib.Path variants, including WindowsPath and PurePath types, are now serialized as plain strings. Pull request #2669
Why it matters
This week brings a new release with Agent Sandboxes, plus changes that protect data: AWS session tokens are now scrubbed and binary attributes survive the scrubber intact. Instrumentation gained more complete recording for streaming errors and for custom and audio responses, and fixes landed for Windows token saving and duplicate Loguru output.
Changelog entry
- Release v5.4.0 Release
- Scrub AWS security tokens by default Pull request #2692
- Carry variable overrides and targeting context into thread pools Pull request #2680
- Drop all incoming context when distributed tracing is disabled Pull request #2679
- Record errors raised during LLM stream consumption Pull request #2491
- Save a read token on Windows where os.fchmod does not exist Pull request #2696
- Add
console_logoption to avoid duplicate Loguru console output Pull request #2690 - Harden remote strings in log messages Pull request #2654
- Keep binary attributes out of the secret scrubber Pull request #2668
- Serialize all pathlib.Path variants as plain strings Pull request #2669
- Treat unbalanced braces as a formatting failure Pull request #2675
- Log dataclass types as their repr, not their default field values Pull request #2649
Logfire 5.4.0 is out, with logfire.agent_sandboxes() and fixes for path serialization, the console exporter and the secret scrubber. Also this week: AWS security tokens are now scrubbed by default, binary attributes are no longer corrupted, thread pools keep variable overrides, a
Logfire 5.4.0 is out. It adds logfire.agent_sandboxes() and includes fixes for pathlib path serialization, empty set formatting in the console exporter, binary attributes in the secret scrubber and exceptions whose str raises. Elsewhere this week, AWS session tokens (X-Amz-Security-Token and aws_security_token) are now scrubbed by default, and disabling distributed tracing drops all incoming context, including baggage. Thread pools now carry managed variable overrides and the targeting context, streaming LLM errors are recorded on the response log, and the read-tokens create command with save works on Windows. A new console_log option prevents duplicate output when Loguru is used alongside Logfire.
Week of October 19, 2026
What shipped
- Released v5.3.0, which automatically instruments FastAPI, Starlette and Flask apps under logfire run and raises when SQLAlchemy instrumentation is skipped. Release
- Added a
pending_spansoption so spans can be sent only once finished, reducing the volume of data sent to Logfire. Pull request #2580 - Added default scrubbing for passphrase,
signing_keyandhmac_keyattributes so they are no longer exported unredacted. Pull request #2593 - Documented an experimental guide for streaming CloudWatch logs and metrics to Logfire through Amazon Data Firehose. Pull request #2589
- Required certifi 2026.10.19 so installing or upgrading the SDK also upgrades an older certificate bundle. Pull request #2640
- Made SQLAlchemy instrumentation raise when it is skipped instead of failing silently. Pull request #2602
Changelog entry
- Release v5.3.0 with automatic FastAPI, Starlette and Flask instrumentation under logfire run. Release
- Add
pending_spansoption to send only finished spans. Pull request #2580 - Scrub passphrase,
signing_keyandhmac_keyby default. Pull request #2593 - Record Responses API refusals as text in
gen_ai.output.messages. Pull request #2618 - Raise when SQLAlchemy instrumentation is skipped. Pull request #2602
- Require certifi 2026.10.19 in SDK dependencies. Pull request #2640
- Document CloudWatch logs and metrics onboarding through Firehose. Pull request #2589
Week of October 12, 2026
What shipped
- Version v5.2.0 was released, adding support for OpenTelemetry 1.45, allowing console attributes without verbose output, adding a compression option with experimental zstd support, and updating FastAPI instrumentation to prevent duplicate telemetry. Release
- OAuth-style token names such as
access_token,refresh_token,id_tokenandbearer_tokenare now scrubbed by default. Pull request #2551 - The console exporter now formats range attributes instead of raising an Unknown data type range error and dropping the argument details. Pull request #2534
- A logfire.tags value passed through the logging and loguru handlers is now recorded as log tags, the same way as tags passed to logfire.info. Pull request #2550
- Url_from_eval links now include the dataset when the report has one, so the compare page opens the right dataset instead of an empty one. Pull request #2566
- Docs were updated to match current feature stages and plan gates, including Gateway Guardrails now in Beta on every Gateway plan and frontend observability and session replay moving to the Settings Early access toggle. Pull request #2567
- The Kubernetes Logs and Events tabs for infrastructure logs are described in the docs as generally available rather than Early access. Pull request #2579
- Docs now state that text search skips words of 40 bytes or longer, so a search for part of a long hash or API key can miss a record. Pull request #2581
Changelog entry
- Release v5.2.0 with OpenTelemetry 1.45 support, console attributes without verbose output, a compression option with experimental zstd support, and a FastAPI instrumentation fix for duplicate telemetry. Release
- Scrub
access_token,refresh_token,id_tokenandbearer_tokenby default. Pull request #2551 - Keep the parent context for spans started while instrumentation is suppressed. Pull request #2570
- Format range attributes in the console exporter instead of raising an assertion error. Pull request #2534
- Treat logfire.tags passed through logging and loguru as log tags. Pull request #2550
- Include the dataset in
url_from_evallinks when the report has one. Pull request #2566 - Update docs to match current feature stages and plan gates. Pull request #2567
- Document Kubernetes infrastructure logs as generally available. Pull request #2579
- Document that text search skips words of 40 bytes or longer. Pull request #2581
- Add OpenTelemetry Collector conformance tests for traces, metrics, logs and TLS. Pull request #2322
V5.2.0 is out: OpenTelemetry 1.45 support, console attributes without verbose output, a compression option with experimental zstd, and FastAPI instrumentation that no longer emits duplicate telemetry. Also this week: OAuth-style token names are now scrubbed by default.
Week of October 5, 2026
What shipped
- Version v6.0.0b8 shipped with support for OpenTelemetry 1.45, an optional compression setting with experimental zstd support, console attributes shown without verbose output, and FastAPI instrumentation that no longer produces duplicate telemetry. Release
- The client gained a compression option, with experimental zstd support that the ingest endpoints already handle, which helps with large spans. Pull request #2535
- Metric counters and histograms now honor an explicitly passed context instead of always using the active span. Pull request #2532
- The disk retryer no longer treats non-retryable responses such as a 401 as success, so queued exports are no longer silently dropped. Pull request #2529
- Raw-byte OTLP posts sent through the session request path now go through the normal retry handling and the disk retryer hand-off. Pull request #2503
- Integers below the int64 minimum are now converted to strings before export instead of being rejected by the encoder and dropped. Pull request #2502
- Fields declared with repr set to false are no longer exported or given nested schemas, matching how dataclasses already behaved. Pull request #2518
- Calling update name on a wrapped span now keeps the span message in sync with the new name. Pull request #2514
- HTTP span messages are now built from the stable HTTP semantic conventions as well as the older attribute names. Pull request #2516
- The command line interface now falls back to the environment base URL setting when no base URL argument is given. Pull request #2260
Why it matters
This week brings a new beta with support for a newer OpenTelemetry release and a compression option that helps with large spans, alongside fixes that keep telemetry complete. Several changes prevent silent data loss, including retries for raw byte exports, correct handling of very small integers, and not deleting queued exports after permanent errors. Issues with duplicate traces, metrics ignoring an explicit context, and spans whose message drifted from their name are also addressed, so the data reaching your backend is more accurate.
Changelog entry
- Release v6.0.0b8 with OpenTelemetry 1.45 support, console attributes without verbose output, compression with experimental zstd, and no duplicate FastAPI telemetry. Release
- Bump the version to v6.0.0b8. Pull request #2542
- Add a compression option and experimental zstd support for exports. Pull request #2535
- Route raw-byte OTLP posts through the existing retry path and disk retryer hand-off. Pull request #2503
- Record metric counters and histograms on an explicitly passed context. Pull request #2532
- Stop treating non-retryable disk retryer responses as success so queued exports are kept. Pull request #2529
- Skip attributes fields with repr set to false when exporting, matching dataclass behavior. Pull request #2518
- Convert span attribute integers below the int64 minimum to strings. Pull request #2502
- Build HTTP span messages from stable HTTP semantic conventions as well as the older names. Pull request #2516
- Keep the span message in sync with the name after update name. Pull request #2514
- Allow console attributes to be shown without verbose output. Pull request #2488
- Default the command line base URL to the environment base URL setting. Pull request #2260
V6.0.0b8 is out: OpenTelemetry 1.45 support, a compression option with experimental zstd, console attributes without verbose output, and no more duplicate FastAPI telemetry. We also fixed dropped retries, lost queued exports and misplaced metrics.
We shipped v6.0.0b8 this week. It adds support for OpenTelemetry 1.45, a compression option with experimental zstd support for large spans, console attributes that can be shown without verbose output, and FastAPI instrumentation that no longer creates duplicate telemetry. Alongside the release, several fixes should make exported data more reliable: raw-byte OTLP posts now go through the retry path, the disk retryer no longer deletes queued exports after permanent errors such as a 401, metrics respect an explicitly passed context, update name keeps the span message in sync, integers below the int64 minimum are converted rather than dropped, and fields with repr set to false are no longer exported. The command line interface also now falls back to the environment base URL setting.
Week of September 28, 2026
What shipped
- Query limits are now documented in a reference page showing limits by organization plan across web UI, read tokens, MCP, and the public API. Pull request #2498
- OpenTelemetry eBPF Instrumentation guide added with setup instructions for Docker Compose and Kubernetes, including OTLP/HTTP export, endpoint configuration, and verification steps. Pull request #2414
- AI Gateway guides published covering setup, provider configuration, routing, protections, optimizations, and spending controls. Pull request #2492
- Documentation clarified the distinction between human annotations from reviewers and end-user feedback scores from live evaluation. Pull request #2504
- TypeScript documentation navigation now mounted directly from the logfire-js repository instead of duplicated in Logfire docs. Pull request #2496
- Documentation updated explaining how cases are ingested into hosted datasets, correcting outdated concepts from earlier docs restructuring. Pull request #2268
- Gateway guides now link directly to corresponding Logfire setup pages for Connect, Providers, Endpoints, Guardrails, and related features. Pull request #2495
- Pyodide dependencies for OpenTelemetry API, instrumentation, and semantic conventions aligned with existing SDK and exporter bounds. Pull request #2493
Why it matters
This week brings documentation improvements that help users understand query limits, set up complex instrumentation scenarios like eBPF and AI Gateway, and navigate between related products. Updates also clarify evaluation concepts and clean up outdated documentation from earlier restructuring.
Changelog entry
- Added query limits reference page documenting limits by organization plan for web UI, read tokens, MCP, and public API Pull request #2498
- Added OpenTelemetry eBPF Instrumentation guide with Docker Compose and Kubernetes setup, OTLP/HTTP export, and verification steps Pull request #2414
- Added AI Gateway guides for setup, provider configuration, routing, protections, optimizations, and spending controls Pull request #2492
- Clarified documentation distinguishing human annotations from trusted reviewers against rubrics from end-user feedback scores Pull request #2504
- TypeScript documentation navigation now included from logfire-js repository with unified page ordering and aliases Pull request #2496
- Updated documentation on how cases are ingested into hosted datasets with corrected product information Pull request #2268
- Added direct links from Gateway guides to Logfire Connect, Providers, Endpoints, Guardrails, and related pages Pull request #2495
- Aligned Pyodide OpenTelemetry dependencies for core API, instrumentation, and semantic conventions with SDK and exporter bounds Pull request #2493
New this week: query limits documented by plan, OpenTelemetry eBPF setup guides, AI Gateway controls walkthrough, and TypeScript docs mounted from logfire-js.
Logfire docs expanded this week with query limit references, comprehensive OpenTelemetry eBPF Instrumentation guides for Docker and Kubernetes, task-focused AI Gateway setup covering providers, routing, protections, and spending controls. TypeScript documentation is now maintained directly in logfire-js and mounted in the main docs.
Week of September 21, 2026
What shipped
- V6.0.0b7 released with a fix to configure proxy managers before other threads can use them. Release
- V5.1.1 released with fixes for project setup on accounts without organizations and TCP keepalive improvements. Release
- V6.0.0b6 released making the logfire package own the logfire command and requiring logfire-cli 0.1.8 or later. Release
- TCP keepalive and idle connection recycling enabled in Logfire's HTTP sessions to prevent stalls from silently dropped connections. Pull request #2309
- V6.0.0b4 released with instrument_mcp() support for mcp 2 and fastmcp 4, and added logfire.instrument_monty(). Release
- Documentation updated to clarify that custom Gateway provider endpoints require a paid plan. Pull request #2484
- Dashboards guide rewritten with updated screenshots reflecting new design and changed features. Pull request #2460
- Coverage reporting moved from Codecov to python-coverage-comment-action matching Pydantic's setup. Pull request #2485
Why it matters
Three beta releases landed in the v6.0.0 line with fixes for proxy configuration and MCP instrumentation, while v5.1.1 brings connection stability improvements to the maintenance branch. Documentation updates clarify plan requirements and reflect current UI behavior.
Changelog entry
- Configure proxy managers before other threads can use them to avoid race conditions Release
- Handle project setup for accounts without organizations Release
- Enable TCP keepalive and recycle idle connections in HTTP sessions Release
- Make logfire package own logfire command and require logfire-cli 0.1.8 or later Release
- Add instrument_mcp() support for mcp 2 and fastmcp 4 Release
- Add logfire.instrument_monty() Release
Logfire v6.0.0b7, v6.0.0b6, and v5.1.1 released. Beta updates fix proxy manager race conditions and clarify command ownership. v5.1.1 improves HTTP connection stability with TCP keepalive and idle recycling.
This week saw three releases across the Logfire lines. v6.0.0b7 and v6.0.0b6 beta releases addressed proxy configuration race conditions and clarified the logfire command ownership with logfire-cli. v5.1.1 on the maintenance branch brings TCP keepalive and connection recycling to prevent stalls from silently dropped connections. Documentation updates clarify custom Gateway provider plan requirements and reflect current UI behavior in dashboards and service levels guides.
Week of September 14, 2026
What shipped
- Logfire 6 split into separate logfire-sdk and logfire-cli packages with the logfire package installing both for backward compatibility. Pull request #2409
- Released Logfire 6.0.0b1 with the SDK and CLI separation, preserving existing imports and optional extras. Release
- Released Logfire 6.0.0b3 requiring logfire-cli 0.1.5 or later. Release
- Added logfire.instrument_snowflake() to instrument Snowflake connector queries with span per query including SQL text, query ID, row count, and connection identity. Pull request #2348
- Fixed logfire.instrument_mcp() to work with MCP 2 and fastmcp 4 by updating module imports. Pull request #2439
- Logfire run now supports executing installed Python
console_scriptsentry points like pytest and uvicorn in instrumented processes. Pull request #2427 - Updated public quickstart documentation to use uvx logfire-cli setup as the primary command. Pull request #2435
- Added documentation for passwordless email-code sign-in and two-factor behavior. Pull request #2440
- Documented public conversation sharing in Early Access including security boundaries and span exposure options. Pull request #2445
- Embedded tutorial videos for Services, Session Replays, and Frontend guides using Cloudflare Stream. Pull request #2436
Why it matters
Logfire 6.0.0 is now in beta with the architecture split into separate SDK and CLI packages, new database instrumentation for Snowflake, MCP compatibility fixes, and expanded documentation covering email authentication, public sharing, and setup workflows. This week consolidates the beta releases and addresses core integration compatibility while improving getting-started guidance.
Changelog entry
- Split logfire into logfire-sdk and logfire-cli with logfire package installing both Pull request #2409
- Released v6.0.0b3 with logfire-cli 0.1.5 requirement Release
- Released v6.0.0b2 with logfire-cli 0.1.4 requirement and console scripts support Release
- Added instrument_snowflake() for Snowflake connector instrumentation Pull request #2348
- Fixed instrument_mcp() compatibility with MCP 2 and fastmcp 4 Pull request #2439
- Logfire run now executes installed Python
console_scriptsentry points Pull request #2427 - Updated browser instrumentation to use configureFrontend for React and Next.js Pull request #2425
- Updated quickstart to use uvx logfire-cli setup as primary command Pull request #2435
- Added documentation for email-code sign-in and two-factor authentication Pull request #2440
- Documented public conversation sharing and security boundaries Pull request #2445
- Reorganized administration documentation with task-oriented management groups Pull request #2431
- Clarified Personal plan billing excludes overage charges unlike paid plans Pull request #2420
Logfire 6.0.0b3 is out. The new split architecture makes logfire-cli independent, Snowflake instrumentation is here, and logfire run now handles console scripts. Try the uvx setup command in the updated docs.
Logfire 6 beta continues to evolve. This week brings the third beta release with logfire-cli 0.1.5, new Snowflake query instrumentation, MCP protocol compatibility fixes, and enhanced setup documentation featuring email sign-in and passwordless authentication. The architecture now cleanly separates the Python SDK from the CLI while maintaining full backward compatibility.