Databases and Data Backends Release Notes
Updates for database engines, hosted databases, and developer platforms built around data storage and retrieval.
Products (8)
Latest Databases and Data Backends Updates
- Nov 10, 2026
- Date parsed from source:Nov 10, 2026
- First seen by Releasebot:Nov 11, 2026
Correlated Logs
Supabase releases correlated logs in Beta, linking events by request, trace, and user so teams can trace activity across Studio and Auth. CLI v3 also brings interactive hosted log inspection, filtering, and JSON output to the terminal.
What's new
Correlated Logs is now available in Beta. The Studio Logs tab now lets you connect log events using three meaningful identifiers: requestId, traceId, and userId.
Every log item now has four tabs: a structured Overview, Raw JSON, and two contextual tabs that appear when relevant:
- Timeline — where a requestId is present, fetches all log items for that request and shows them as a timeline. You can select any item in the chain to see its details and navigate back to the full list.
- User — where a log item is associated with an authenticated user, shows their details and links directly to that user in Auth. From the Auth user page, you can view all their logs in the same unified view.
You can also filter the logs list by user attributes to see all log items for a specific user, in the same way you filter by requestId or traceId.
For AI agents using the Supabase MCP server, the query_logs tool supports the same identifiers. See MCP debugging.
How to use it
Open the Logs tab in your project dashboard and filter by requestId, traceId, or user attributes. Click into any log item to see the Timeline or User tab where available.
To send a traceId from your application, see client-side tracing.
Logs in CLI v3
Inspect hosted logs in your terminal:
supabase logs --remote --service postgres --level error --since 30mFilter by service, severity, or time range. Use --from and --to for fixed UTC bounds.
In an interactive terminal, use the up and down arrows to select a row, Enter to inspect it, C to copy the full record as JSON, and L to load newer logs on demand.
For scripts and agents, --output-format json and --output-format stream-json read once and exit. Query hosted logs with --sql or load SQL with --file.
The command defaults to local logs, which require the experimental stack runtime.
Read the CLI logs guide for setup and examples.
Why we built this
When something goes wrong, the signal is spread across log items from different services, and tracking it back to a specific request or user means cross-referencing timestamps by hand. These connections should be a click away.
This is a Beta release. Try it and share feedback in the GitHub Discussion below.
Original source - November 2026
- Date parsed from source:Nov 8, 2026
- First seen by Releasebot:Nov 11, 2026
pg_pinecone
Pinecone adds hosted-model text embedding in pg_pinecone v0.12.1, enabling meaning-based search without a vector column.
pg_pinecone can now embed a table’s text column with a hosted model, such as gemini-embedding-2, so you can search by meaning without a vector column. It’s available in v0.12.1. To upgrade an existing install, see the operator runbook.
Original source All of your release notes in one feed
Join Releasebot and get updates from Supabase and hundreds of other software products.
- Nov 6, 2026
- Date parsed from source:Nov 6, 2026
- First seen by Releasebot:Nov 7, 2026
Breaking Change: supabase-flutter v3 brings type safety to the whole API surface
Supabase releases supabase-flutter v3 as a major update, now available in dev prereleases, with end-to-end typed database APIs, Dart-native data types, and coding-agent migration skills. The additive typed API supports gradual adoption, while v2 receives critical fixes for six months after stable launch.
supabase-flutter v3 is the next major version of the Supabase client libraries for Flutter and Dart. You can try it today through the 3.0.0-dev prereleases on pub.dev. The stable 3.0.0 release is planned for November 11, 2026.
What you get
Types across the whole client.
supabase gen types --lang dartturns your database schema into a Dart file with a row, insert, and update type for every table. The newsupabase.table(Books.table)API uses them end to end: selects returnList<BooksRow>instead ofList<Map<String, dynamic>>, a filter such asBooks.price.lt(20)only accepts a value of the column's type, writes take generated insert and update types, embedded selects follow foreign keys, andstream()emits typed rows. Postgres functions get typed methods onsupabase.rpc. Beyond the database, the connection state of the realtime client, channel results, AuthState, and every service exception are typed too. A renamed column breaks your build, not your users' sessions.An API that reads like Dart. Timestamps are
DateTime, URLs areUri, durations areDuration, realtime listeners are Streams, AuthState is a sealed class you can switch over exhaustively, and every service exception extends one SupabaseException base.Skills for coding agents. The supabase, supabase_flutter, supabase_typegen, and supabase_test packages ship package skills for Claude Code, Cursor, Gemini, and Copilot. They teach your agent the v3 API instead of the v2 code it was trained on, and include a skill that carries a v2 project through the upgrade.
The typed API is additive. from() keeps working on the same client, so you can move one table at a time. It is marked @experimental while its shape settles.
Why v3 breaks v2 code
Large parts of v2 were ported directly from supabase-js. Timestamps and URLs were strings, realtime used callbacks, and order() sorted descending by default, the opposite of SQL, with nothing in the analyzer or the results to reveal it. A major version is the only moment these shapes and defaults can change, so v3 changes all of them at once. You upgrade once instead of absorbing a series of smaller breaks.
Who is affected
Every Dart and Flutter project on supabase_flutter or supabase v2 that upgrades to v3.
Projects that depend directly on gotrue, realtime_client, storage_client, or functions_client. These are renamed to supabase_auth, supabase_realtime, supabase_storage, and supabase_functions, and the old packages are marked discontinued on pub.dev when v3 ships. If you only depend on supabase_flutter or supabase, your dependencies and imports stay the same.
Projects that use a third-party package built on supabase_flutter v2. You can upgrade once that package releases a v3-compatible version.
Self-hosted projects running a GoTrue older than API version 2024-01-01. v3 only reads the newer error codes.
What happens if you take no action
Nothing changes. A ^2.x version constraint never resolves 3.0.0, and pub never picks a prerelease unless you ask for one. v2 keeps receiving security and critical fixes for 6 months after the stable 3.0.0 release. New features only land in v3.
Changes the compiler will not catch
Most of the upgrade shows up as compile errors. Some changes do not, such as the default sort direction of order() and how apps on v2 and v3 exchange realtime broadcasts, so check the full list of changes the compiler will not catch in the migration guide before you ship.
Migration steps
Read the changes the compiler will not catch. Add ascending: false to every order() call that relies on descending order, while you are still on v2.
Bump supabase_flutter or supabase to ^3.0.0 in pubspec.yaml. Replace direct dependencies on gotrue, realtime_client, storage_client, and functions_client with their renamed packages.
Install the package skills, for example with
dart run skills@ get -p supabase_flutter, then ask your coding agent to run the migration skill if you work with one.Fix the compile errors with the migration guide. For example, anonKey on Supabase.initialize becomes publishableKey. Your existing anon key is still a valid value.
Go through the changes the compiler will not catch one by one, and release every app that shares a realtime broadcast channel at the same time.
Optionally, add supabase_typegen as a dev dependency, generate your types with Supabase CLI 2.122.0 or newer, and move your queries from from() to table().
Rollout timeline
Date Milestone User action Now v3 available as 3.0.0-dev prereleases on pub.dev Optional: try v3 and report issues on GitHub 2026-11-11 3.0.0 stable. gotrue, realtime_client, storage_client, and functions_client are marked discontinued on pub.dev Upgrade by following the migration steps 6 months after the stable release v2 stops receiving security and critical fixes Finish the upgrade to v3Questions or problems with the upgrade? Open an issue on supabase-flutter or contact Supabase support.
Original source - Nov 6, 2026
- Date parsed from source:Nov 6, 2026
- First seen by Releasebot:Nov 7, 2026
NOV 06, 2026
Neon adds AI Gateway credit purchases on Free plans, Console-based project recovery, automatic updates for computes up to 16 CU, new MCP and coding-agent support, fresh Functions templates, and sorted live queries in Realtime SDK 0.2.0.
All accounts can purchase AI Gateway credits
The AI Gateway lets you add AI features to your app, like summarizing notes, describing uploaded photos, or answering customer questions. One Neon credential gives you more than 60 models from OpenAI, Google, and other providers. You don't need a paid plan to use it anymore. On the Free plan, an organization admin can buy prepaid credits from $5 under Billing in the Neon Console. You pay only for the tokens you use, at each model's rate in the model catalog. To start building, see Get started with the AI Gateway.
Recover deleted projects from the Console
You can now recover a deleted project from the Neon Console, in addition to the CLI and API. Recovery is available for 7 days after deletion. On your Projects page, select Manage on the banner, then Recover. See Recover a deleted project.
Automatic updates for computes up to 16 CU
If you've been manually restarting computes between 8 and 16 CU to keep them current, you can stop. Every compute up to 16 CU, including all autoscaling computes, now gets compute updates automatically. On paid plans, they run in the update window you choose.
An update restarts the compute, which typically takes a few seconds. Your data isn't touched, and Neon prewarms the cache during the update, so queries don't run cold afterward. Fixed-size computes above 16 CU still need a manual restart. To learn more, see Updates.
Give your app its own MCP server
Your customers already use AI agents like Claude, Cursor, and ChatGPT. Now they can use your app from them too. Run an MCP server on Functions and turn on the OAuth provider plugin in Managed Better Auth. Your users sign in with the account they already have, and the agent can only reach their own data. Start from the template:
neon functions new mcp-postgresTo add sign-in and per-user data, see Protect an MCP server with the OAuth provider plugin. For the backstory, read the blog post.
Set up Junie, Kiro CLI, Pi, and more coding agents with Neon
One command connects your coding agent to Neon, with the Neon MCP Server and agent skills set up for you. Factory Droid, fx, Hermes Agent, Junie, Kilo Code, Kimi Code, Kiro CLI, and Pi are newly supported this week. Run:
npx neon@latest initTo set up one agent, pass --agent , like --agent junie. Neon works with more than 20 coding agents, including Claude Code, Cursor, Codex, and VS Code. See the full list of supported agents.
New Neon Functions templates
We added new templates to the Function Registry this week:
- AI agents and chat: chat-sdk, mastra, openai-agents, tanstack-ai
- APIs: openapi-postgres, rate-limit-postgres, x402-paid-api (pay per request)
- Workflows: inngest-workflow, upstash-workflow
- Email: aiinbx (contributed by AI Inbx), resend-email
- Storage, realtime, and monitoring: presigned-uploads, websocket-live-poll, sentry
List them and add one to your project from the Neon CLI:
neon functions templates list neon functions new openai-agentsNew version of the Neon Realtime SDK
Version 0.2.0 of the Neon Realtime SDK is out. Realtime (beta) turns any SQL query into a live query, so your app's UI updates the moment your data changes. Upgrade for the latest improvements, including sorted live queries that stay sorted as rows change. To upgrade, update the SDK and any adapters you use at the same time:
npm install @neon/[email protected] @neon/[email protected]If you're new to Realtime, start with the Realtime quickstart.
Fixes & improvements
- Read replicas
- Neon CLI
- Managed Better Auth
- Agent skills
- Neon Management SDK
- Neon MCP Server
- AI Gateway
- Nov 6, 2026
- Date parsed from source:Nov 6, 2026
- First seen by Releasebot:Nov 7, 2026
Webinar Recap: From Data to Vectors to Context
Pinecone introduces Nexus, a knowledge engine for AI agents that adds task-specific context, source citations and rules for resolving conflicting information, helping agents make more reliable decisions from retrieved data.
What AI Agents Need Beyond RAG
Ask an agent for projected Q4 mobile growth. It generates SQL, runs it and comes back with +16%. But September 8 was a one-day partner promotion, and the agent counted the spike as demand. Exclude the promotion from the forecast, keep it in reported totals, and the same trend gives +6%. Those are illustrative values. The data was all there and the SQL was valid.
In a webinar on November 4, Jörg Schad, Pinecone's VP of Engineering, traced retrieval from keyword search to RAG, covered what changes once agents consume it, and took questions, including how to make semantic search more accurate. That question usually has its answer outside the search.
This post was adapted from Jörg Schad's webinar From Data to Vectors to Context. Here's the full replay:
What Similarity Can't Tell You
Imagine the two nearest neighbors for a question about a feature are its design doc from three years ago and a scan of the codebase from this morning. Both are similar to the query. Only one describes the code that runs today, and a similarity score can't tell you which.
Each era of retrieval fixed the limit of the one before it. Keyword search runs on an inverted index and ranks with relevance metrics (IDF dates to 1972, BM25 to 1994), which gives exact match at scale. But "cancel" never matches "terminated subscription."
Semantic search closed that gap. With word2vec (2013), HNSW for approximate nearest-neighbor search (2016) and BERT (2018), retrieval finds related meaning across vocabulary. Its limit is the design-doc problem. Similar isn't authoritative.
RAG, from Lewis et al. in 2020, gave models a way to ground answers in a knowledge base at query time, and it has been in nearly every chat demo since. The grounding is unchecked, though. The model gets whatever comes back, with no step that checks whether it is the right result or whether the model used it.
Why Good Recall Still Gives Wrong Answers
Retrieval quality is usually measured as recall at K. Recall@10 is the share of relevant documents that land in the top 10 results. It's a good measure of whether the information is there at all.
Now take a policy that a newer one has superseded. Retrieval returns both versions, and recall counts both as relevant. A plausible answer grounded in the old version scores zero, and Recall@K can look fine anyway. The same goes for a document the caller isn't allowed to see: relevant to the query, and out of bounds for the answer.
So how do you make semantic search more accurate? Often the search is already doing its job. After a merger, one company's fiscal year runs January to December and the other's runs February to January, so the same quarter covers different months on each side. For any question that spans both, semantic search finds every relevant document. It can't say how to combine them.
Tuning the retriever won't help there. No similarity score carries a rule for lining up two fiscal calendars.
Retrieval is scored on Recall@K: did the right documents come back? Context is scored on the decision made with them. Every example so far splits the same way:
Query What retrieval returns What decides it Projected Q4 mobile growth Complete data, including a September 8 spike That September 8 was a one-day partner promotion, excluded from the forecast How does this feature work? A three-year-old design doc and this morning's code scan Which source describes the code that runs today Any figure spanning both companies after a merger Every relevant document from both How a January-December fiscal year lines up with a February-January one What does the policy allow? The current policy and the version it superseded Which version is in forceWhy It Matters More When an Agent Reads the Answer
A person who sees a forecast far too high, or dropping when it shouldn't, raises an eyebrow and double-checks. An autonomous agent might ship it without that check, and agents call retrieval at volumes no human search ever generated.
A high-volume agent may not notice that results don't make sense or are out of date, and it may not ground the answer on its own. It's missing the context that would tell it.
So why not give the model everything and let it work out what's relevant? Because more input tends to make answers worse.
Chroma's context rot report evaluated 18 LLMs and found that performance grows less reliable as input length grows, even on deliberately simple tasks. On LongMemEval, focused prompts averaging about 300 tokens beat full prompts averaging around 113,000 tokens. The same holds for agents: load one with information that isn't specific to its task and the outcomes get worse.
The question changes. Whether the right documents came back still matters, but an agent also needs to know what is valid, trusted, permitted and appropriate to use for this task.
Where the Deciding Happens
Search stays. Different tasks need different retrieval tools, and plenty need more than one. We put dense vectors, sparse vectors and full-text search with BM25 ranking in one index. Keyword search still earns its place wherever exact match matters, and all three are built into the database rather than beside it.
The deciding layer sits on top of search. Each task gets a task context: summaries, entities, policies and rules, plus knowledge artifacts curated in the background before any query arrives. The context resolves which artifacts are relevant to the task, then issues lexical, vector or hybrid queries against the index as it needs them.
Take the merger. Search already found every document. The missing piece is semantic information: how one fiscal calendar maps onto the other, held in the task context next to the policies and rules. The superseded policy works the same way, and so does the databot's one-day promotion. The index supplies the data, and a rule in the context decides how to treat it.
We built Pinecone Nexus, our knowledge engine for agents, to be that layer. Nexus curation keeps a ledger of processed sources, tracked by content hash, so a run only touches what changed. Answers carry citations to the file, page and verbatim span behind each claim.
Precedence gets written down, too. A context's default instructions act as a system prompt for every query against the context, and they can tell the model how to handle conflicting sources, such as which policy version wins.
We don't think Nexus is the answer for every workload. If your retrieval patterns are simple, you're fairly sure the top 10 results hold the answer, no complex rules or policies decide which source wins, and you're fine owning the grounding, RAG is enough.
Operational data that changes every second, like each purchase on a large online store, probably wants a live, dynamic view. Real-time feeds and records edited in place all day aren't a good fit for Nexus today. It works best where data changes on a slower schedule, such as contracts and research.
Before you spend a sprint on embeddings or reranking, check whether retrieval already found the documents. If it did, the fix belongs to the layer that decides. The full replay is on YouTube. To build that layer, start with the Nexus docs or request a trial.
Original source - Nov 6, 2026
- Date parsed from source:Nov 6, 2026
- First seen by Releasebot:Nov 7, 2026
Nov 06, 2026 – AI Gateway credits for all accounts, project recovery in the Console, automatic updates up to 16 CU, and more
Neon expands AI Gateway access to Free accounts, adds Console recovery for deleted projects and automatic updates for computes up to 16 CU, and ships new agent integrations, Functions templates, Realtime improvements, and extensive CLI, auth, SDK, and MCP updates.
All accounts can purchase AI Gateway credits
The AI Gateway lets you add AI features to your app, like summarizing notes, describing uploaded photos, or answering customer questions. One Neon credential gives you more than 60 models from OpenAI, Google, and other providers. You don't need a paid plan to use it anymore. On the Free plan, an organization admin can buy prepaid credits from $5 under Billing in the Neon Console. You pay only for the tokens you use, at each model's rate in the model catalog. To start building, see Get started with the AI Gateway.
Recover deleted projects from the Console
You can now recover a deleted project from the Neon Console, in addition to the CLI and API. Recovery is available for 7 days after deletion. On your Projects page, select Manage on the banner, then Recover. See Recover a deleted project.
Automatic updates for computes up to 16 CU
If you've been manually restarting computes between 8 and 16 CU to keep them current, you can stop. Every compute up to 16 CU, including all autoscaling computes, now gets compute updates automatically. On paid plans, they run in the update window you choose.
An update restarts the compute, which typically takes a few seconds. Your data isn't touched, and Neon prewarms the cache during the update, so queries don't run cold afterward. Fixed-size computes above 16 CU still need a manual restart. To learn more, see Updates.
Give your app its own MCP server
Your customers already use AI agents like Claude, Cursor, and ChatGPT. Now they can use your app from them too. Run an MCP server on Functions and turn on the OAuth provider plugin in Managed Better Auth. Your users sign in with the account they already have, and the agent can only reach their own data. Start from the template:
neon functions new mcp-postgresTo add sign-in and per-user data, see Protect an MCP server with the OAuth provider plugin. For the backstory, read the blog post.
Set up Junie, Kiro CLI, Pi, and more coding agents with Neon
One command connects your coding agent to Neon, with the Neon MCP Server and agent skills set up for you. Factory Droid, fx, Hermes Agent, Junie, Kilo Code, Kimi Code, Kiro CLI, and Pi are newly supported this week. Run:
npx neon@latest initTo set up one agent, pass --agent , like --agent junie. Neon works with more than 20 coding agents, including Claude Code, Cursor, Codex, and VS Code. See the full list of supported agents.
New Neon Functions templates
We added new templates to the Function Registry this week:
- AI agents and chat: chat-sdk, mastra, openai-agents, tanstack-ai
- APIs: openapi-postgres, rate-limit-postgres, x402-paid-api (pay per request)
- Workflows: inngest-workflow, upstash-workflow
- Email: aiinbx (contributed by AI Inbx), resend-email
- Storage, realtime, and monitoring: presigned-uploads, websocket-live-poll, sentry
List them and add one to your project from the Neon CLI:
neon functions templates list neon functions new openai-agentsNew version of the Neon Realtime SDK
Version 0.2.0 of the Neon Realtime SDK is out. Realtime (beta) turns any SQL query into a live query, so your app's UI updates the moment your data changes. Upgrade for the latest improvements, including sorted live queries that stay sorted as rows change. To upgrade, update the SDK and any adapters you use at the same time:
npm install @neon/[email protected] @neon/[email protected]If you're new to Realtime, start with the Realtime quickstart.
Fixes & improvements
Read replicas
Paid plans now support up to 6 read replicas per branch. Branches that already have more keep them, but can't add another until they're under the limit. The Free plan limit of 3 per project hasn't changed.
Neon CLI
Update to the latest version to get these changes. See Upgrade the Neon CLI.
New neon postgres command (aliases pg and lakebase) groups the Postgres commands and adds computes, so you can list, inspect, resize, start, suspend, restart, and delete computes from the CLI.
neon orgs now manages organizations: get, members (list, get, update role, remove), invitations (list, create), and spending-limit.
neon roles create --access read_only or --access read_write creates a least-privilege role instead of an admin. Add --database and --schema to scope it (the schema defaults to public). The same access option is on neon.postgres.roles.create in @neon/sdk 7.8.0 and on the MCP Server's create_postgres_role tool, so an agent asked for a read-only credential no longer hands out an admin. Without access, roles are still created as admins. See neon roles create.
New neon roles get and neon roles reset-password read a single role and rotate its password.
New neon operations get shows the status, duration, and any error for a single operation.
neon bootstrap now offers to deploy at the end when the template's environment variables are all set. Pass --deploy to install, link, and deploy without prompts.
neon claim create and neon init --claimable now accept --project-name, so a claimable project keeps the name you choose after it's claimed. See neon claim.
neon init now finishes linking your project, writing neon.ts, and pulling environment variables even when plugin, skills, or MCP setup fails for one of your agents.
neon plugins and neon init now check that GitHub Copilot CLI and Grok Build actually run before installing the Neon plugin through them.
neon init --services data-api and neon config init --services data-api no longer declare Managed Better Auth for a Data API that verifies tokens from an external identity provider such as Clerk or Auth0. They now read how the linked branch's Data API is set up.
neon databases delete and neon roles delete now tell you when the database or role is already gone instead of printing nothing, and failed commands no longer print the update notice.
Commands for Realtime, Functions, function triggers, Object Storage, and the AI Gateway now explain when a feature isn't available in your project's region. The error names the project's region and the regions where the feature runs, instead of a bare 404 (@neon/sdk 7.7.0). See Regions.
A top-level realtime key in neon.ts now fails with a message that tells you to configure Realtime under preview: { realtime: true }.
neon plugins and neon init now always install the Neon plugin user-level. This fixes project-level installs, which failed for every agent in the previous release. VS Code, GitHub Copilot CLI, Grok Build, and Kimi Code no longer need --global, and --global is still accepted, so existing scripts keep working.
Managed Better Auth
Managed Better Auth now runs Better Auth 1.7.16, which fixes the issues in Better Auth's November 3 security release. The fix is server-side and already live in every region, so you don't need to do anything: no SDK upgrades, no secrets to rotate.
A new Revoke sessions on password reset setting signs a user out of every existing session after a successful password reset, whether they reset through a link or Email OTP. It's off by default and set per branch. Turn it on in Settings > Better Auth under Sign-in with Email, with neon neon-auth config email-password update --revoke-sessions-on-password-reset, or with the Neon API. See Sign users out after a reset.
@neon/auth 0.8.1 works with Better Auth UI 1.7. Thrown errors keep the raw error code, password reset and email verification links no longer redirect signed-out users to sign-in, and auth.middleware() accepts publicRoutes for pages that don't need a session.
Agent skills
The Neon agent skills now cover branching and egress optimization inside neon-postgres, so the standalone neon-postgres-branches and neon-postgres-egress-optimizer skills are retired. If you installed either by name, install neon-postgres instead. The skills also teach Upstash for caching, rate limits, and queues, and the Managed Better Auth OAuth server for MCP and OAuth clients. To install or update the skills, run npx neon@latest skills.
Neon Management SDK
@neon/sdk adds a neon.organizations namespace for reading an organization and managing its members, invitations, spending limit, and API keys. Every method defaults to the client's orgId.
neon.auth now covers the branch-level Managed Better Auth settings, including localhost access, email and password sign-in, the email provider, the magic link, organization, and phone number plugins, and webhooks.
Neon MCP Server
The hosted Neon MCP Server now supports the MCP 2026-07-28 specification (MCP 2.0), and existing clients keep working unchanged. The old HTTP+SSE endpoints, /sse and /message, are retired and return 410 Gone. If your client still points at /sse, change it to https://mcp.neon.tech/mcp. See Retired HTTP+SSE transport.
New tools: count_projects answers "how many projects do I have", and get_project_consumption and get_branch_consumption report usage from the consumption history API.
Agents can now change Managed Better Auth branch settings, not just read them: localhost access, email and password sign-in, the email provider (with a test send), the organization, magic link, and phone number plugins, and webhooks. get_neon_auth_config also reports the new revoke_sessions_on_password_reset setting.
MCP clients and directories now show the server as Neon, with its logo, description, and docs link. The server also publishes a Server Card at https://mcp.neon.tech/.well-known/mcp.json for discovery before sign-in.
The search and fetch tools now work for personal accounts that belong to several organizations. Previously, search failed with a "Multiple organizations found" error. Thanks to breken-ai for finding and diagnosing the issue.
Realtime, Functions, trigger, Object Storage, and AI Gateway tools now return the same region-aware error as the CLI when a feature isn't available in your project's region.
AI Gateway
@neon/ai-sdk-provider 1.3.0 uses gpt-image-2 for image generation by default, since OpenAI deprecated gpt-image-1. If you call neon.tools.imageGeneration() without a model, upgrade or set model: 'gpt-image-2' to keep it working. Requests that still name gpt-image-1 are rejected before they're sent. See Image generation with Vercel AI SDK.
The model catalog now shows each model's lab, plus the context window and sub-cent prices for decision models.
Original source - November 2026
- Date parsed from source:Nov 4, 2026
- First seen by Releasebot:Nov 11, 2026
Twice the included embedding tokens through December 31
Pinecone doubles monthly embedding token allowances for Starter and Builder organizations through December 31, 2026, with other limits unchanged.
From November 4 through December 31, 2026, Starter and Builder organizations get twice the included tokens on every embedding model. Starter includes 10,000,000 tokens per month per model (5,000,000 before the promotion), and Builder includes 20,000,000 (10,000,000 before). Reranking allowances and per-minute rate limits don’t change. Allowances return to their usual levels on January 1, 2027.
For the current limits, see Rate limits.
Original source - Nov 4, 2026
- Date parsed from source:Nov 4, 2026
- First seen by Releasebot:Nov 5, 2026
Breaking Change: Supabase CLI v3 is local by default and confirms hosted changes
Supabase releases CLI v3 with safer, consistent defaults: commands target the local stack unless directed to a hosted project, hosted changes require confirmation, and configuration follows a unified precedence. The major upgrade also changes schema diff behavior and removes deprecated commands, requiring CI and deployment updates.
Supabase CLI v3.0.0 is the next major version of the Supabase CLI. npm, Homebrew, Scoop, the Linux packages, and supabase/setup-cli now install it by default. It changes which project a command runs against, when a command asks for confirmation, and where configuration values come from. Most pipelines that deploy with the CLI need a small change before they upgrade.
What changed
Database commands target your local stack by default. db push, db pull, db dump, migration list, gen types, and the storage commands run against the local stack unless you pass --remote. One --remote flag replaces --linked, --project-ref, and --db-url, which still work but are deprecated.
Every command that changes a hosted project asks for confirmation first, and the default answer is No. When the CLI can't ask, as in CI, the command fails unless you pass --yes. Piping y into a command no longer confirms it.
Flags and environment variables override config.toml, including [remotes.] blocks. Every value resolves in the same order: flag, shell environment, project env files, config file, then default.
pg-delta is the default schema diff engine, so the SQL that db diff and db pull generate can change.
Deprecated commands, flags, config keys, and environment variables are removed. A removed command or flag exits with code 1 and prints its replacement. This includes the [inbucket] config section, which is now [local_smtp].
The CLI ships as a single supabase executable. The separate supabase-go executable is no longer installed.
Two smaller changes can affect a production project. config push now applies [auth.sms] enable_confirmations as written. secrets unset defaults to No and needs --all to remove every secret.
Why we made this change
In v2, some database commands defaulted to the linked project and others to the local stack, and four different flags chose a target. A command run without a flag could reach a production database. Many hosted changes ran without a confirmation, or with one that defaulted to Yes. Configuration had the same problem: config.toml beat environment variables for some settings and not for others.
v3 gives every command the same rules. Commands run locally unless you say otherwise, ask before any hosted change, and read every setting in one order. Defaults like these can only change in a major version, so v3 changes them together.
Who is affected
Anyone who runs the CLI in CI, in deploy scripts, or through an AI coding agent. Hosted changes that run unattended now need --yes, and database commands need --remote to reach your hosted project.
Projects created with supabase init before v2.108.0. Their supabase/config.toml has an [inbucket] section, so supabase start, stop, status, and the db commands fail until you rename it.
Projects that deploy with [remotes.] blocks and also set SUPABASE_* environment variables. The variables now take precedence over the block, and config push pushes them.
Projects that edit supabase/schemas and then run db diff. Under pg-delta, use supabase db schema declarative sync instead.
Scripts that call removed commands or flags, or call supabase-go directly.
v3 changes the CLI only. Hosted projects, their data, and self-hosted Supabase are not affected.
What happens if you take no action
Nothing changes until you install v3. A pipeline that installs the latest CLI without pinning a version already runs v3. The most common failures are:
- supabase start fails with [inbucket] in supabase/config.toml is no longer supported.
- functions deploy, secrets set, config push, and db push fail in CI with HostedMutationConsentRequiredError.
- db push without --remote targets the local stack. It fails when no local stack is running. If your job started one, it succeeds there and never reaches your hosted project.
Migration steps
Update to the latest v2 release and run your usual workflows, including CI. v2.122.0 and later warn about removed commands, flags, and config keys, and about commands whose default target changes in v3. Fix each warning.
Rename [inbucket] to [local_smtp] in supabase/config.toml, along with any [remotes..inbucket] block. The keys inside don't change. v2.108.0 and later already read [local_smtp], so you can do this before you upgrade.
Add --remote to every command that should reach your hosted project, and --yes to every hosted change that runs unattended. The quick reference for CI maps common v2 commands to their v3 form.
If you use [remotes.] blocks, check the SUPABASE_* variables in CI and in your env files. Run supabase config diff before your first config push on v3.
If supabase/config.toml has no project_id, run supabase stop before you upgrade. Local container and volume names can change in v3.
Install v3 and work through the rest of the migration guide.
To upgrade later, pin v2 with npm install [email protected] --save-dev, brew install supabase/tap/supabase@2, scoop install supabase-v2-stable, or version: v2.stable on supabase/setup-cli@v3.
Rollout timeline
Date Milestone User action 2026-11-02 v3.0.0 released. npm, Scoop, the Linux packages, and supabase/setup-cli install v3 by default. v2.122.1 adds the v2.stable channels. Follow the migration steps, or pin v2 2026-11-03 brew install supabase installs v3 Same as above Ongoing v2 receives security fixes and fixes for broken behavior, but no new features Upgrade to v3 to get new featuresIf v3 breaks a workflow the migration guide doesn't cover, run supabase issue bug or open an issue.
Original source - November 2026
- No date parsed from source.
- First seen by Releasebot:Nov 3, 2026
New embedding models for document indexes
Pinecone adds hosted Google and OpenAI embedding models for document indexes, with flexible dimensions, support for up to 8,192 tokens, and task-specific embedding options. Availability varies by plan, with US processing available for select models and plans.
Document indexes can now embed text with three more hosted models: gemini-embedding-2 from Google, and text-embedding-3-small and text-embedding-3-large from OpenAI. Each takes any dimension in a range, up to 3,072, and embeds up to 8,192 tokens per value.
gemini-embedding-2 embeds text for a task type you choose, such as search, question answering, or clustering. You can also override the query task for one search with parameters on an embed clause. It’s available on the Builder, Standard, and Enterprise plans.
The text-embedding-3 models are available on every plan. On the Standard and Enterprise plans, you can keep their processing in the US with processing_location. Processing locations lists where every hosted model processes text.
Original source - Nov 1, 2026
- Date parsed from source:Nov 1, 2026
- First seen by Releasebot:Nov 11, 2026
Redis Enterprise Software by Redis
7.8.6-346 (November 2026)
Redis Enterprise Software fixes a security vulnerability and improves support package generation, alongside internal fixes.
Security fix for CVE-2026-111497.
Bug fix for support package generation.
Internal fixes and improvements.
Original source