Highlights
Platform architecture and core services
The initial architecture comes together across the release: an etcd-backed configuration manager (#11 — @khushijainslack, later generalized to any key-value store in #500 — @abhishekgml1991), Kafka-driven async record processing (#83), ArangoDB graph schema and record permissions (#8 — @rohilbansal-pipeshub), and internal service routes with token-based service auth (#23, #27). The services tree was restructured into thebackend/ layout that persists today (#423 — @abhishekgml1991), and the connectors service moved to its long-term port 8088 (#252).
RAG pipeline: from first Chat API to token-level streaming
The query service grew from an initial Chat API (#33 — @abhishekgml1991) into a full retrieval pipeline: query rewrite and expansion with async API calls (#66), multi-question query decomposition (#74), a reranker (#96), LangGraph orchestration (#277), and follow-up query transformation (#420). Answers stream token-by-token end to end — streaming API (#431), model-native streaming (#456), citation-safe chunking (#457), and frontend streaming support (#451 — @vishwjeet-pawar03). Normalized citations with web URLs (#438, #360) and a low-latency Quick mode as the default (#458, #645) round out the pipeline.Bring-your-own-model: 10+ LLM and embedding providers
Model support expanded from OpenAI-only to a broad provider matrix: Gemini (#114), Anthropic (#120 — @abhishek-pipeshub), AWS Bedrock (#128, #506), Ollama for local models (#313, #475), any OpenAI-compatible endpoint (#346), and xAI, Mistral, Cohere, Together, Groq, and Fireworks in one refactor that made adding new providers trivial (#483, #484 — @abhishekgml1991). Embeddings got the same treatment: Cohere and Gemini (#338), Ollama (#480), OpenAI-compatible APIs (#413), and Voyage, JinaAI, and Mistral (#485).Connector fleet and data-source SDK
First-class connectors arrived for Google Workspace (Drive and Gmail with webhooks, watches, and resync — #250 — @rohilbansal-pipeshub), OneDrive (#594), SharePoint Online (#635), Confluence Cloud (#785), Jira (#542), Slack (#546 — @kaushal1999), Outlook (#802 — @shrikantNemiwal1), Dropbox (#808 — @R0gue-one), and web pages (#928). Underneath, an OpenAPI-driven client generator (#580, #584 — @rish231294) and a connector registry (#653 — @vishwjeet-pawar03) enabled a wave of community-built data sources: GitHub (#771 — @uncleSlayer), GitLab (#795), Notion (#663), S3 (#694), Azure Blob (#697), Box (#703), Linear (#740), Airtable (#746), Zendesk (#751), Discord (#815 — @MinuteHanD), Freshdesk (#821), LinkedIn (#826 — @darshangodase), PagerDuty (#912), BookStack (#823), ServiceNow (#829), Evernote (#852), Asana (#834), and Zammad (#902), plus a connector integration playbook (#850 — @shrikantNemiwal1) and generic OAuth support (#396 — @ShawnSack).Blocks-based document parsing and indexing
Indexing converged on a unified blocks schema (#406, #648 — @kaushal1999) with format-specific converters for DOCX (#679), Excel (#684), Gmail (#698), and Slack mrkdwn (#553). Image indexing and multimodal support landed (#863, #688 — @vishwjeet-pawar03), Docling handles complex PDFs with thread-safety and slowness workarounds (#909, #739), vector points are quantized to Uint8 to cut Qdrant memory (#468), and records stuck in-progress after a crash are recovered (#814).Agents and enterprise tools
Beyond RAG, the release seeds the agent runtime: agent tools (#522 — @rish231294), Google enterprise tools (#535), a no-code agent builder (#600 — @vishwjeet-pawar03), agentic RAG streaming (#463), agent citation handling (#791 — @kaushal1999), and improved tool calling (#845).Knowledge bases, search UI, and auth
Users get per-org knowledge bases with multi-KB support (#286, #487 — @vishwjeet-pawar03), in-app viewers with citation highlighting for PDF, DOCX, Excel, HTML, text, and Markdown (#154, #168), knowledge search with filters (#334), dark mode (#364, #417), and document summaries (#268 — @rohilbansal-pipeshub). On the auth side: SAML SSO (#105 — @khushijainslack), JWT secret handling (#144), secret key hashing (#262), Microsoft MSAL login fixes (#558), and a run of dependency security patches (multer, samlify, pdf.js, xlsx, Vite — #388, #389, #259, #274, #256).Developer experience & infrastructure
- Production Docker Compose with simplified config, model downloads baked into images, and slim Python base images (#178, #290, #143, #918, #474).
- PR validation via GitHub Actions with Node.js and Python linting, typed-Python enforcement, and CodeQL scanning (#237, #238, #371, #654).
- Community scaffolding: contribution guidelines, code of conduct, security policy, issue and PR templates (#289, #612, #778, #752, #779).
- Swagger/OpenAPI docs for storage APIs, Prometheus metrics collection, feature flags, and PostHog analytics (#93, #246, #865, #839 — @jatingaur18).
New contributors
- @abhishekgml1991 made their first contribution in #1
- @vishwjeet-pawar03 made their first contribution in #6
- @rohilbansal-pipeshub made their first contribution in #8
- @rish231294 made their first contribution in #16
- @abhishek-pipeshub made their first contribution in #81
- @dependabot[bot] made their first contribution in #388
- @ShawnSack made their first contribution in #396
- @kaushal1999 made their first contribution in #489
- @ParmodBisht made their first contribution in #637
- @rish664 made their first contribution in #652
- @uncleSlayer made their first contribution in #771
- @MinuteHanD made their first contribution in #815
- @darshangodase made their first contribution in #826
- @shrikantNemiwal1 made their first contribution in #850
- @jatingaur18 made their first contribution in #839
- @R0gue-one made their first contribution in #808
Full commit-level detail: GitHub release · Commit history