2026-05-15
- More capable agents — New tools for graphs, embeddings, document parsing, markdown conversion, and table extraction; no hard cap on agent turns.
- Bird Cage execution — Replaces the previous sandbox with a faster, more reliable environment for running agent code.
- MCP integrations (beta) — Connect rooms and agents to external tools via MCP, with in-app setup for room-level integrations.
- Sharing improvements — Revoke shared threads, working charts on shared links, and better previews for PDFs, CSVs, and large messages.
- Billing & usage — New usage dashboard, invoice listing, and current-cycle usage; fixes for trials and billing edge cases.
- Skills per agent — Enable or disable individual skills for each agent.
- Auth & security — Login via backend, WorkOS widget, finer-grained access tokens, dedicated user-content hosting, and ongoing security hardening.
- Agent v5 (internal) — Redesigned agent UI with clearer progress, inline links and images in replies, and a smoother chat experience.
2026-04-24
- Billing and plans — Stripe billing enabled end-to-end, with clearer trials, seat-based pricing, and safer plan changes.
- More AI models — Anthropic Claude models available, alongside an improved default model lineup.
- Smarter agents — More reliable answers after tool use, optional web search by default, and a refreshed “thought process” experience.
- Share conversations — Share chat threads with branded link previews and working charts.
- Easier onboarding — New org onboarding flow and fixes for sign-in and org creation.
- Behind the scenes — A new agent proxy layer improves how artifacts and data are served to the UI.
2026-04-15
- Billing foundation — Developer plan added, Stripe enabled in dev, storage accounting for usage, and groundwork for end-to-end billing.
- Streaming thought process — Agent reasoning streams live in the UI, with fixes when multiple tables appear in the right panel.
- Feature view downloads — Download links for feature views directly in the app.
- Smarter table previews — Agents can preview a single row from very large tables without loading the full dataset.
- Clearer agent output — Tool results are grouped so you see the latest output; fixes for tool calls and large-table handling.
- Platform upgrades — Python 3.14, event log service, and reliability fixes for storage accounting and tracked tables.
2026-04-10
- Billing service — New billing APIs, subscription cancellation, Stripe meter sync, and local billing fallback.
- Room & thread skills — Room-level skills with publish/demote to threads; updated skills, artifacts, and user-files UI.
- External agent tools — Configure which tools each agent can use, with UI for external tool integrations.
- Smarter agents — Per-message time limits, better web-search prompts, context-window fixes, and GPT 5.4 support.
- Canvas & charts — Auto-layout on the canvas, edge rendering fixes, and Vega charts upgraded to react-vega v8.
- Data sources — Wired to the new data-source service; file-upload fixes for resource tables.
- Downloads & previews — Feature view and file download URLs; non-UTF-8 CSV handling.
- Platform & reliability — A2A agent HTTP surface, vector-search improvements, multi-token-issuer auth, and monitoring on AWS/Azure.
2026-03-27
- Skills — Agents can use skills (read, write, edit, and view files, upload skills) with a new skills UI.
- Web search & enrichment — Web search tool for agents, web sources in the app, and Web Augment for data enrichment.
- Agent v3 — V3 is the default experience; artifact listing in the side panel, query interrupt, and streaming/markdown fixes.
- MCP for agents — MCP support for v3 agents, with MCP URLs shown in the agent integrations tab.
- Stripe integration — Initial Stripe billing integration on the backend.
- Chat improvements — Preview context files in threads, delete messages, and list artifacts from the side panel.
- Team settings — Change member roles from the settings page; “Profile” renamed to “Settings”.
2026-03-16
- Agent v3 & canvas (beta) — Major V3 rollout: new canvas with auto-layout, Graph Builder, Join, Augment, and Embed operations; output detail pages; rename outputs; and broad UI polish.
- Agent Skills — Backend skills tools (read, write, edit, view files) and skills UI on the frontend.
- Chat & context files — Upload files to chat threads, preview context files, and notifications when file processing completes.
- Agent capabilities — Citations in Agent v3, query interrupt, exposed tool names/arguments, and Gemini 3.1 on the frontend.
- Agents & threads — Add, clone, and delete agents; threads on agents; connect agents to workflows.
- Auth & admin — Auth0 removed in favor of WorkOS; audit log tab and WorkOS audit export; change member roles.
- Platform — Billing/management service split, Python 3.13, LiteLLM router with caching, and infrastructure updates.
2026-02-24
- Database ingestion — Ingest database tables (with Airbyte worker support on the backend).
- Snowflake connector — Snowflake data connector in the UI.
- Cloud ingestion — Enabled everywhere; MCP support for deployed agents and cloud-ingestion syncing.
- Citations — Citation tags beside images in agent replies and broader citation rendering.
- Auth & backoffice — Auth0 removed; WorkOS APIs moved to org service; backoffice member panel and invite flows.
- Platform — LiteLLM in agents, Deno sandbox with expanded Python/scientific libs, and infrastructure updates (Next.js 16, Python 3.13).
- Corvic V3 UI (internal) — New V3 design end-to-end, with canvas nodes and edges, improved styling, and streaming agent responses.
2026-01-28
- Snap Actions UI — Snap Action and Snap Feature View flows in the app, with clearer distinction between normal and snap actions.
- MCP for deployed agents — Agent ingress for MCP endpoints, MCP syncing quota, and fixes for agent sync and republish.
- Sandbox in production — Sandbox service deployed on Azure and GCP for isolated code execution.
- Feature views — Error states, images rendered in table cells, and loading states on details.
- Agent experience — Auto-scroll in reasoning, open context from any node, improved reflection UI, and better markdown rendering.
- LLM Augment — Markdown columns treated as file prompts, clearer failure reasons, and duplicate output-name prevention.
- Platform — Gemini thinking level, local access-token creation, and documentation links pointed to docs.corvic.ai.
2026-01-08
- Snap Actions — New Snap Actions capability for faster, guided workflows.
- Sandbox execution — Dedicated sandbox service for code augment and safer, isolated code execution.
- MCP for agents — Deploy agents as MCP endpoints, sync cloud ingestions and workflows, and check whether an agent has digested resources.
- Smarter orchestration — Tool-calling orchestrator enabled across deployments, with improved reflection prompts.
- Ingestion & storage — Storage quota checks on blob ingestion and auto-generated document titles.
- Auth cleanup — Auth0 variables removed; WorkOS-only auth path.
- Reliability — Agent metering fixes, faster quota checks, Vega v6 upgrade, and broad stability fixes.
2025-11-24
- Introducing Workflows! A new way to build workflows, explore data, and chat with agents.
2025-10-17
- Reflection mode now supported in agents.
2025-09-24
- Citations in deployed agents.
- Quota and plans.
2025-09-19
- Streamable HTTP for MCP.
2025-09-02
- Fixed an issue ingesting Parquet files.
- Fixed an issue ingesting PDFs with many images on a page.
2025-08-27
- Fixed an issue where image descriptions did not match their images.
2025-08-16
Ingestion
- Faster processing with support for larger PDF files.
- Incrementally add data from external connectors like Amazon S3 and Azure Blob Storage.
- Upload CSV data to unstructured pipelines.
- Tabular embedding spaces now accept user input columns.
- Plots in agent messages can now be customised.
- Additional operations (Data Analysis, Find Relevant Documents) added to support CSV data.
2025-07-17
- Upload data from external data connectors like Amazon S3 and Azure Blob Storage.
- Admins can configure external data sources (Amazon S3 buckets, Azure Blob Storage) from the dashboard.
2025-07-09
- Upload Text, Markdown, HTML, and DOCX files in addition to PDF/PPTX for unstructured pipelines.
- New image embedding model: SigLIP2, a multilingual vision-language encoder by Google DeepMind.
- New agent prompt types: Persona Prompt (tone and style) and Workflow Prompt (task-specific guidance).
- New Chain of Actions operations: Retrieve Adjacent (one-hop neighbor retrieval) and Temporal Analysis (trends and seasonality).
- SQL queries available in agent context, and AI-generated starter queries for deployed agents.
2025-06-11
- Fixed an issue with a deployed agent being locked before deployment.
- Disabled deployment preview during config updates to avoid race conditions.
- Fixed broken images and links in agent messages.
- Added support for citations of images in preview.
2025-05-23
- Search and filter by status across multiple PDF files in a single pipeline.
- Clone space configuration and set custom space names.
- Deployed agents (beta) — integrate an MCP endpoint to use Corvic agents in your system.
- Agent citations (beta) — answers include references to source documents.
2025-05-18
- Upload PPTX files in addition to PDF for unstructured pipelines.
- Advanced user management — organize and invite members to your organization.
- Enhanced, more secure sign-in experience.
- More accurate AI insights, clone agent configuration, and improved tablet/iPad UX.
2025-04-14
- File download and preview options; improved tracking for uploads and failures.
- Custom LLM integrations — bring your own endpoints (OpenAI Generic and Azure OpenAI compatible).
- Subscription plans — Premium and Enterprise tiers introduced.
- Usage dashboard — track data and computation usage across rooms.
- Live context for agents, custom query suggestions, and new completion LLMs (Gemini 2.0 Flash).
2025-03-10
- Ingestion — unstructured (PDF: documents, images, slides) and structured (Parquet) data, complex transformations, multi-file and incremental uploads.
- Embeddings — semantic (Vertex AI, OpenAI), image (CLIP), tabular, and graph-structure embeddings.
- Exploration — multi-modal exploration, combined image/text search, advanced filters, and export.
- Agents — multi-space, multi-modal chat, dynamic chain of actions, and major completion LLMs (GPT-4o, Gemini 1.5).
2025-02-27
- Vector search with back-linking to original data.
- Access control and interactive visualizations.
- Improved structural transformation engine (text-to-graph, table-to-text, graph-to-text).
- Out-of-the-box LLM offerings.
- Generation of embedding pipelines through the UI.
- Upload tabular data in Parquet format.
- Autonomous table-to-graph transformation and Node2Vec graph embeddings.
- Embedding visualization via UMAP, quality metrics, and Parquet export.