Skip to content

docs: finalize v1.4.0 changelog #29

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
MUDASSIR-75 merged 1 commit into from
Oct 22, 2026
Merged

docs: finalize v1.4.0 changelog #29

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 9 additions & 6 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,9 +2,8 @@

All notable changes to this project will be documented in this file.

## [1.4.0] — In development
## [1.4.0] — `2026-10-6`

Version metadata aligned to `1.4.0` on 6 October 2026. Release date, tag and publication remain pending.

### ⚠️ Compatibility

Expand All @@ -20,15 +19,13 @@ Version metadata aligned to `1.4.0` on 6 October 2026. Release date, tag and pub
### 📖 Documentation & Interfaces

- Remove the internal `testing-folder/LONGTRAINER_ROADMAP.md` from the repository; public configuration guidance remains in the named-agent guide.

- Add optional `agent_type` to the existing HTTP build request; invalid names return HTTP 400 with supported names.
- Add `longtrainer build [BOT_ID] --agent-type ...` and purpose selection on `bot create`, preserving existing CLI options and explicit empty tool overrides.
- Register the named-agent guide, document infrastructure-only YAML and interface precedence, and consolidate duplicate migration navigation entries.

### 🧪 Testing & CI

- Install the parser model explicitly in CI rather than relying on a transitive spaCy dependency; keep all Python matrix jobs running when one fails.

- Collect the full `tests/` suite on Python 3.11 and 3.12, including rate-limiter tests and 17 relocated bug-fix regressions.
- Make legacy script checks and lazy-loading integration setup failures fail pytest; reject tests returning non-`None` values.
- Add 16 offline lazy-loading cases, two failure-signaling regressions, and a separate MongoDB/in-memory Qdrant CI job with an `integration` dependency extra.
Expand All @@ -42,31 +39,34 @@ Version metadata aligned to `1.4.0` on 6 October 2026. Release date, tag and pub
- Identify Wikipedia requests with a project user agent so live lookups satisfy the service's request policy.
- Fall back to Yahoo Finance's news-search headlines, including source links and publication timestamps, when ticker article lookup returns no news.
- Parse provider text blocks into strings for structured and vision responses and agent invocation and synchronous/asynchronous streaming, including Gemini responses.

- Extract reusable JSON-schema validation, deterministic schema hashing, concise error feedback, and one-retry response repair into `longtrainer.structured`.
- Preserve structured chat dictionaries and successful-only history updates; copy retry messages, remove dead message construction and the temporary Ruff lambda exception, and honor invocation configuration on both attempts.
- Add structured bot/chat/API boundary regressions; retain unused schema registry methods without adding persistence side effects.

## [1.3.1] — 2026-05-07

### ✨ New Features

- **Tenant Rate Limiting**: Two-layer execution rate limiting system using a token-bucket algorithm. Layer 1 enforces per-tenant RPM ceilings, Layer 2 enforces equal-share budgets across bots within a tenant. Fully configurable via `longtrainer.yaml` or `RateLimitConfig`.
- **Per-Bot Overrides**: Fine-grained rate limit overrides for specific bots via `bot_overrides` configuration.
- **CLI Rate Limit Recovery**: The `longtrainer chat` and `longtrainer add-doc` commands now auto-retry with a progress bar when rate limits are hit.
- **API 429 Responses**: FastAPI server returns proper `429 Too Many Requests` with `Retry-After` headers when rate limits are exceeded.

### 🔧 Improvements

- **Interactive Demo**: Added `demos/longtrainer_demo.py` — a complete RAG workflow demo with live Rich progress tracking. Fully env-var driven (supports OpenAI, Gemini, Ollama, any vector store) with zero hardcoded config.
- **CLI Cleanup**: Reverted CLI from Rich TUI panels to clean `click.echo/secho` output. Removed `rich>=13.0` from core dependencies.
- **Removed VHS/TUI assets**: Cleaned up `.tape` files, `Makefile`, and demo GIFs from the repository.

### 📖 Documentation

- New `rate_limiting.md` guide covering two-layer enforcement, YAML configuration, API error handling, and CLI auto-retry.
- Updated README with embedded demo screenshot and streamlined Quick Start section.

## [1.3.0] — 2026-05-01

### ✨ New Features

- **LongTracer Integration**: Native, optional support for LongTracer hallucination detection and observability.
- Automatically captures spans, latencies, and token counts for LLM calls and retriever queries.
- Provides claim verification via `CitationVerifier` using hybrid STS and NLI for RAG, Vision, and Structured bot responses.
Expand All @@ -76,27 +76,30 @@ Version metadata aligned to `1.4.0` on 6 October 2026. Release date, tag and pub
## [1.2.3] — 2026-04-29

### 🚀 Performance & Scalability

- **Lazy Loading Chat Histories:** Significantly optimized startup time and RAM consumption. `load_bot()` no longer eagerly loads all chat histories into RAM. Instead, chat histories are lazy-loaded on-demand from MongoDB via new `_ensure_chat_loaded()` and `_ensure_vision_chat_loaded()` methods.
- **Qdrant Initialization Fix:** Fixed a bug in `vectorstores.py` where passing `:memory:` to Qdrant would fail due to improper URL parsing. In-memory fallback is now properly handled via the `location` parameter.

## [1.2.2] — 2026-04-11

### 🐛 Bug Fixes & Perf Improvements

- **Async Document Ingestion:** Implemented `aadd_document_from_path` and `aadd_document_from_link` with parallel ingestion via `ThreadPoolExecutor` and `asyncio.gather()`. Batch document ingestion is now non-blocking and significantly faster.
- **Retrieval Confidence Scores:** Added `invoke_vectorstore_with_scores()` which returns the raw FAISS L2 distance and automatically injects a normalized `retrieval_score` (1.0 = perfect match) into document metadata.
- **KeyError Bug:** Fixed a crash during `update_chatbot()` caused by an obsolete `faiss_path` dictionary reference.
- **Documentation:** Added a new comprehensive implementation roadmap `/testing-folder/LONGTRAINER_ROADMAP.md` covering Phase 3 through Phase 10.


## [1.1.0] — 2026-02-21

### ✨ New Features

- **Zero-Code CLI**: Introduced `longtrainer` command-line interface. Use `longtrainer init` for interactive project scaffolding (generates `longtrainer.yaml`) and `longtrainer serve` to start the REST API.
- **FastAPI REST Server**: Shipped a comprehensive built-in API server (`longtrainer.api:app`) with 16 endpoints covering bot management, document ingestion, sync/streaming chat, vision chat, and vector search.
- **Lazy Initialization**: Trainer instance now defers LLM connection initialization until the first API call, allowing the server to start successfully (`/health` check passes) without an `OPENAI_API_KEY`.
- **API Documentation**: Auto-generated Swagger UI available at `/docs` when serving the API.

### 📦 Dependencies

- Added `click>=8.0` and `pyyaml>=6.0` to core dependencies for CLI and configuration management.
- Added `[cli]` optional dependency group (`click`, `pyyaml`).
- Automatically installs `fastapi` and `uvicorn` when using the `[api]` extra.
Expand Down
Loading