Releases: Schromeo/SledTrace
Release list
SledTrace v0.7.0 — External Developer Readiness
SledTrace v0.7.0 — External Developer Readiness
SledTrace v0.7.0 makes the existing local-first RAG debugging stack easier for an external developer to evaluate, install, validate, and contribute to. It is a readiness and distribution release, not a new diagnostic-feature release and not a claim of external adoption.
Highlights
- independent GitHub Actions jobs for Python 3.9, Python 3.13, the Go Collector, and the Dashboard
- protected
mainbranch with those four required checks - OIDC Trusted Publishing workflow for TestPyPI and production PyPI
- TestPyPI
0.7.0rc1publication and clean-index validation - first production PyPI release for ordinary
pip install sledtrace - clean-clone startup instructions with Docker and explicit non-Docker prerequisites
- deterministic reference-trace and Dashboard validation evidence
- current SledTrace-branded README screenshots
- focused contribution templates and a repeatable release checklist
Install the Python SDK
python -m pip install sledtrace==0.7.0Preferred usage remains:
from sledtrace import traceThe legacy raglens import remains temporarily available for migration compatibility.
Serving boundary
The wheel contains the Python SDK and sledtrace CLI. It does not bundle the Go Collector, Dashboard, Docker files, or a standalone serving runtime.
sledtrace --help,sledtrace serve --help, andsledtrace versionwork after package installation.sledtrace servestarts local services only from inside a SledTrace source checkout.- Outside a checkout,
sledtrace serveexits non-zero with guidance to clone the repository and use the documented local stack.
Clean-clone evidence
The v0.7 release candidate was cloned into a new directory and validated with the non-Docker fallback on Windows:
- Collector health returned
{"service":"sledtrace-collector","status":"ok"}. - all nine deterministic
reference_rag_apptraces flushed successfully - the Dashboard displayed current SledTrace branding and all nine traces
- the conflict case displayed 30-day versus 14-day retrieved-chunk evidence
- the weak case displayed low-score, weak-match, and unsupported-claim diagnostics
Docker Desktop was installed on the validation host but could not start because WSL2 virtualization was disabled. The Docker path therefore remains covered by the existing Docker configuration and prior v0.4 smoke evidence; the v0.7 clean-clone run used the documented non-Docker path and added explicit Docker prerequisites.
Scope preserved
v0.7.0 does not add:
- span types beyond
retrievalandllm - LangChain or LlamaIndex adapters
- cloud hosting, authentication, or hosted services
- agent/tool/memory tracing
- LLM-as-judge evaluation
- new warning rules or storage/API schema changes
Production publication validation
- immutable annotated tag
v0.7.0targets release commit58887907973aff3948d2cf3667681832f4305ec6 - protected Trusted Publishing workflow passed: https://github.com/Schromeo/SledTrace/actions/runs/34410674101
- production package: https://pypi.org/project/sledtrace/0.7.0/
- a no-cache install of
sledtrace==0.7.0from production PyPI succeeded in a new virtual environment outside the repository - preferred
sledtraceand temporary legacyraglensimports passed and reported0.7.0 - installed CLI help, serve help, and version passed
- wheel-installed
sledtrace serveoutside a checkout exited non-zero with the documented source-checkout guidance
Release-preparation validation
- Python SDK: 17 tests passed
- wheel and source distribution: built successfully for
0.7.0 - package metadata and long description:
twine checkpassed - clean-wheel preferred and legacy imports: passed
- installed CLI help, serve help, and version: passed with version
0.7.0 - wheel-installed
serveoutside a checkout: expected non-zero exit and guidance passed - Go Collector API, storage, and warning tests: passed
- Dashboard
0.7.0TypeScript and Vite production build: passed - repository whitespace check: passed
SledTrace v0.6.0 — Local CLI / Startup UX
SledTrace v0.6.0 — Local CLI / Startup UX
Summary
SledTrace v0.6.0 adds a small package-installed CLI and makes its serving boundary explicit. The CLI works after editable or wheel installation, while local service startup remains source-checkout based.
This release does not turn the Python wheel into a standalone SledTrace runtime.
CLI behavior
Available commands:
sledtrace --help
sledtrace serve --help
sledtrace version
sledtrace servesledtrace --helpdescribes the installed CLI.sledtrace serve --helpexplains that serving requires a source checkout.sledtrace versionprints0.6.0.sledtrace servewalks upward from the current working directory to locate a SledTrace checkout.- Inside a checkout,
servedelegates toscripts/start-sledtrace.py. - Outside a checkout,
serveexits non-zero and explains how to run from the repository or use Docker Compose.
The checkout markers are:
AGENTS.mddocker-compose.ymlscripts/start-sledtrace.py
Standalone wheel limitation
The v0.6.0 wheel does not bundle:
- the Go Collector binary
- the Dashboard production build or Node runtime
- Docker images or Compose runtime assets
- platform-specific startup binaries
As a result, normal wheel installation supports CLI help and version reporting, but not standalone serving. Run sledtrace serve from inside a SledTrace source checkout, or use docker compose up --build from the repository root.
Compatibility guarantees
- preferred import remains
sledtrace - temporary
raglensimport compatibility remains available - preferred collector variable remains
SLEDTRACE_COLLECTOR_URL - legacy
RAGLENS_COLLECTOR_URLfallback remains available - the existing repo-local startup script remains the
servebackend - collector API routes and payload contracts are unchanged
- SQLite schema and legacy database compatibility are unchanged
- Dashboard data contract is unchanged
Validation
Validated on 2026-09-07:
cd sdk/python
pytest -q
python -m build
python scripts/validate-wheel.py
cd collector/go
go test ./... -count=1
cd dashboard/web
npm.cmd run buildObserved results:
- Python SDK: 17 tests passed; output included the expected legacy
raglensdeprecation warning and an environment-specific.pytest_cachepermission warning - packaging:
sledtrace-0.6.0.tar.gzandsledtrace-0.6.0-py3-none-any.whlbuilt successfully - clean wheel environment:
import sledtrace,from sledtrace import trace, andimport raglenspassed - clean wheel CLI: root help,
serve --help, andversionpassed - clean wheel
serve: returned the expected non-zero status and documented source-checkout guidance outside a checkout - editable installation: succeeded
- repo detection: located the checkout from a nested repository directory
- repo-local delegation: passed under a subprocess test double, without starting long-running services
- Go Collector: all packages and tests passed
- Dashboard: TypeScript and Vite production build passed
Unchanged product scope
v0.6.0 introduces no new product functionality beyond CLI/startup UX:
- supported span types remain
retrievalandllm - existing deterministic warning rules are unchanged
- no new warning rules or LLM-as-judge behavior
- no collector protocol changes
- no storage schema changes
- no Dashboard data-contract changes
- no LangChain or LlamaIndex adapters
- no cloud, authentication, hosted Collector, or PyPI publication
Release state
Released on 2026-09-08.
- release commit:
392edd1233a99e20f2cf7ffdfa166cdbb689bb6e - annotated tag:
v0.6.0 - GitHub Release: SledTrace v0.6.0 — Local CLI / Startup UX
- PyPI publication: not performed
The release commit also removed tracked Python bytecode/cache artifacts and added repository ignore coverage so routine Python test runs do not pollute the working tree.
SledTrace v0.5.0 — Python SDK Packaging Readiness
SledTrace v0.5.0 — Python SDK Distribution / Packaging Readiness
Summary
SledTrace v0.5.0 focuses on the Python SDK as a clean local package artifact. The goal is to make the SDK installable from a built wheel or source artifact, importable via the preferred sledtrace path, and compatible with temporary legacy raglens imports during migration.
This release is intentionally scoped to packaging readiness and compatibility support. It does not change the warning engine, collector API, storage schema, dashboard data contract, or span-type set.
Status
Released on 2026-09-08.
- release commit:
b3cad60a10636dbf7a5d371f51bac0c04a4af936 - annotated tag:
v0.5.0 - GitHub Release: SledTrace v0.5.0 — Python SDK Packaging Readiness
- PyPI publication: not performed
Highlights
- SledTrace-first package metadata and versioning
- local wheel and sdist build support from
sdk/python - clean public import surface via
from sledtrace import trace - temporary legacy
raglenscompatibility shim - collector URL precedence validation for
SLEDTRACE_COLLECTOR_URLandRAGLENS_COLLECTOR_URL - local wheel validation script for clean install checks
- documentation for source install, wheel build, and local distribution flow
Installation from source
cd sdk/python
pip install -e .Local wheel build
cd sdk/python
python -m pip install --upgrade pip
python -m pip install build
python -m buildThis produces wheel and source-distribution artifacts in sdk/python/dist/.
Install built wheel
pip install dist/*.whlWheel validation
cd sdk/python
python scripts/validate-wheel.pyThe script creates a temporary virtual environment, installs the built wheel, and runs import checks for the supported public import paths.
Compatibility notes
- Preferred public import:
from sledtrace import trace - Temporary compatibility import:
from raglens import trace - Preferred env var:
SLEDTRACE_COLLECTOR_URL - Legacy fallback env var:
RAGLENS_COLLECTOR_URL
Precedence remains:
SLEDTRACE_COLLECTOR_URLRAGLENS_COLLECTOR_URLhttp://localhost:4319
Validation commands
cd sdk/python
python -m pip install --upgrade pip
python -m pip install build
python -m build
python -m venv .venv-package-test
source .venv-package-test/bin/activate
pip install dist/*.whl
python -c "import sledtrace; print(sledtrace.__version__)"
python -c "from sledtrace import trace; print(trace)"
python -c "import raglens; print('legacy raglens import ok')"
deactivate
pytest
cd collector/go
go test ./... -count=1
cd dashboard/web
npm run buildCurrent limitations
- PyPI publishing is not performed by this release.
- No warning engine changes are part of v0.5.0.
- No collector API contract changes are part of v0.5.0.
- No storage schema changes are part of v0.5.0.
- No dashboard data contract changes are part of v0.5.0.
- No new span types are introduced.
- The
raglenspackage remains a temporary compatibility layer only.
Out of scope
- PyPI publishing automation
- hosted/cloud deployment features
- new framework adapters
- LLM-as-judge behavior
- changing the local-first architecture
SledTrace v0.4.1 — Rebrand
RAGLens has been renamed to SledTrace to avoid confusion with an unrelated research project.
This compatibility-preserving release updates the project branding, documentation, startup tooling, package metadata, and public configuration names without changing the core tracing, diagnostics, storage, or dashboard functionality introduced in v0.4.0.
Migration
- Use
SLEDTRACE_COLLECTOR_URLinstead ofRAGLENS_COLLECTOR_URL. - Use
scripts/start-sledtrace.pyinstead ofscripts/start-raglens.py. - Use the
sledtracePython import path for new integrations. - Legacy configuration, launcher, and Python imports remain temporarily supported.
Validation
- Go collector tests passed.
- Python compatibility tests passed.
- Dashboard production build passed.
- Docker Compose configuration and end-to-end reference RAG smoke tests passed.
- Existing SQLite data and public trace APIs remain compatible.
SledTrace continues as a local-first observability and debugging platform for RAG pipelines, with future development expanding toward evaluation and AI workflow harness capabilities.
RAGLens v0.4.0 — Local Release / Install & First-Run Experience
Project renamed: RAGLens is now SledTrace.
See the v0.4.1 release for the current project name and migration details.
RAGLens v0.4.0 — Local Release / Install & First-Run Experience
RAGLens v0.4.0 packages the project as a cleaner local-first release candidate for external developers.
This release focuses on first-run developer experience: Docker-based local startup, explicit configuration, health checks, reference demo validation, local reset guidance, release notes, and improved documentation.
Highlights
- Added Docker Compose local stack for the Go collector and React dashboard.
- Added Dockerfiles for the collector and dashboard.
- Added
.env.examplefor local configuration defaults. - Added
.dockerignoreto keep Docker build context clean. - Added MIT license.
- Added v0.4.0 release notes.
- Added reference RAG app runbook.
- Added optional health-check script.
- Updated README with Docker-first quickstart and fallback local startup path.
- Updated smoke-test and maintainer handoff docs.
- Preserved existing Python SDK trace contract.
- Preserved deterministic-first warning generation.
- No changes to collector API, storage schema, dashboard data contract, or warning rule behavior.
Validation
Validated with the following commands:
cd collector/go
go test ./... -count=1
cd dashboard/web
npm run build
cd ../..
docker compose up --build
curl http://localhost:4319/health
curl http://localhost:5173
cd sdk/python
pip install -e .
python -m examples.reference_rag_app.run allAll expected reference traces were generated:
- reference-rag-app-refund
- reference-rag-app-conflict
- reference-rag-app-wrong-window
- reference-rag-app-processing-range
- reference-rag-app-wrong-processing-range
- reference-rag-app-damaged
- reference-rag-app-digital
- reference-rag-app-subscription
- reference-rag-app-weak
Current Implemented Span Types
- retrieval
- llm
Current Warning Rules
- no_retrieved_chunks
- low_retrieval_score
- duplicate_chunks
- weak_query_chunk_overlap
- numeric_mismatch
- conflicting_chunks
- answer_not_grounded
Not Included in This Release
- PyPI publishing
- LangChain adapter
- LlamaIndex adapter
- hosted collector
- auth
- cloud sync
- agent/tool/memory spans
- LLM-as-judge default evaluator
Recommended Quickstart
From repo root:
docker compose up --buildThen in another terminal:
cd sdk/python
pip install -e .
python -m examples.reference_rag_app.run allOpen:
http://localhost:5173