Release notes for October 2026

Grepr released these feature updates, improvements, and bug fixes in October 2026.

October 24, 2026

Faster backfills of large log volumes

Backfills of large log volumes run faster, and large and unlimited backfills that used to run out of resources or time out now complete. A backfill with a limit no longer stops after sending only part of it. The Jobs page has a new Backfills tab that shows each backfill’s progress through selecting events and sending them, and how many events each vendor received. When you start a backfill, click Track progress in the confirmation message to open it there. See Backfill logs from the data lake to your vendors.

Other improvements and fixes

You can now connect Cursor, Codex, and ChatGPT to the Grepr MCP server with OAuth sign-in. See Connect to the Grepr MCP server.
An agent’s issue settings are on a new Issues step of the agent’s page, and agents search the issue board for issues related to the problem they are investigating instead of starting from a list of known issues. See How an agent finds issues.
Searching the issue board matches words in an issue’s title, description, labels, entities, and recent comments, and lists the best matches first.
Memory tag filters on the Memories page and in agents’ memory searches ignore case, so service:Checkout finds memories tagged service:checkout.
An issue’s comments are listed newest first, directly under the comment box, with Show older comments at the bottom.
The data warehouse CloudFormation template expires the query results Grepr writes to your bucket after one day.
Fixed an issue where an AI agent’s investigation or chat reply could end on interim text, such as “Let me check”, while the agent was still working.
Fixed an issue where live edit could stop applying a pipeline’s masking and reducer patterns to the logs it showed.

October 21, 2026

Pass log events between pipelines with Grepr pipes

You can now create a Grepr pipe on the Integrations page and use it to carry log events from one pipeline to another. A pipe sink in the sending pipeline writes events into the pipe, and a pipe source in the receiving pipeline reads them unchanged.

Track and work issues with AI agents on the issue board

You can now track problems on the issue board, where people and AI agents file, work, and close issues together, and the board refreshes while you watch it. You can hand an open issue to an agent from the issue page, or let an agent pick up issues from the Ready column on its own, filtered by priority, label, entity, and team. When a run ends, the agent posts a comment on the issue that says how it ended. See Work on issues with AI agents.

Other improvements and fixes

You can now add, change, and remove a pipeline’s tags from its details page without restarting the pipeline, and filter the Pipelines page by tag.
You can now define your own mapping from status values to severities on the attributes remapper’s Status Remapping tab. See Define your own status mapping.
A Datadog-like *: search matches its term as a whole word in the message and in every tag and attribute value, the same way a bare term matches the message. See Search logs using a Datadog-like syntax.
You can now cancel every queued, running, and paused investigation for one agent or for your whole organization with Cancel active on the agent’s page or the Agents page, and pause the affected agents in the same step. See Run and review investigations.
The filter step in the log pipeline editor has a redesigned layout. See Filter and transform events with actions.
Searches can be longer and contain more terms, and searches for many values of one field run faster.
Searches that filter on a dotted attribute, such as @http.request.method, run faster.
Pipeline updates are faster when several steps of the pipeline use the same integration or dataset.
You can now chat with the AI assistant on an Anthropic LLM integration as well as a Gemini one, and pick a model tier, such as Opus or Sonnet, instead of a model name. See Configure LLM provider connections with a Grepr integration.
An LLM integration that leaves Max tokens per window empty now allows 200,000,000 tokens per window instead of 20,000,000, and both budget fields are under Advanced on the integration form. See Cap token consumption with a budget window.
The AI assistant opens from a button in the bottom-right corner of the page instead of the app bar.
Fixed an issue where a pipeline stopped for more than a day could fail to start again, and such a pipeline now starts without its saved state, so reducer patterns and in-progress aggregations start over.
Fixed an issue where updating or stopping a pipeline under heavy load failed after five minutes when Grepr could not save its state in time, and the pipeline now resumes from its most recent checkpoint, which might send some logs to your vendor a second time.
Fixed an issue where, in rare circumstances, a pipeline you stopped could keep running.
Fixed an issue where some NRQL-like queries on raw logs failed with an error that a column could not be found.
Fixed an issue where a search that listed several hundred values in one OR list failed to run.
Fixed an issue where logs that an external trigger or the trace sampler marked to skip aggregation were still aggregated in newly created pipelines.
Fixed an issue where, after you ran a search with a relative time range in Log Explorer or Trace Explorer, typing in the search box could hide results that were already loaded.
Fixed an issue where a filter you typed in live edit was saved into a template-based log pipeline when you clicked Apply.
Fixed an issue where Start on a failed or deleted pipeline opened the stop dialog, and where renaming an unsaved pipeline opened the create dialog.
Security fixes.

October 16, 2026

Review and correct what your AI agents remember

You can now see the memories every AI agent in your organization recorded on the new Memories page in the Agents area, filter them by entity tag, agent, significance, verdict, or text, and open one to read it with links to the investigations it came from. With permission to edit integrations, you can edit a memory, mark it wrong with a note, or delete it, and agents treat the root cause of a memory you marked wrong as ruled out. Each agent’s Memories step also sets how much the agent searches its memory and which conclusions it records, and can start each investigation with the memories related to its triggering signals. See Review memories in the UI.

Other improvements and fixes

Attribute remapping rules and the Grok parser’s Extract from field can now name an attribute whose name contains a dot, such as user.agent in {"http": {"user.agent": "curl/8.4.0"}}, when you write the path as a JSON array of attribute names, such as ["http", "user.agent"]. See Write a path as an array of attribute names.
The AI assistant opens from the app bar, its model picker lists only text chat models and supports search, and you can copy a reply.
Grepr now stores a NaN, infinite, or out-of-range number in an incoming log or span as null.
Fixed an issue where a dataset or integration could not be deleted after you deleted a pipeline that used it, when an earlier update of that pipeline had failed.
Fixed an issue where a search too long for the query engine, such as one containing a large pasted block of text, failed with an internal error that repeated the whole query. Log Explorer, Trace Explorer, Analyze, agent investigations, and the Grepr CLI now report that the query is too large so you can simplify it.
Fixed an issue where a Datadog Agent payload that was within the size limit when compressed but too large after decompression, such as a large trace payload, failed with a 500 error instead of a 413 error.

October 14, 2026

Ask an assistant about your logs and pipelines

You can now hold an ongoing conversation with a Grepr AI assistant from a dockable panel in the product. The assistant searches your logs, inspects your pipelines, and can apply changes for you, and it asks before any change takes effect, so you can allow it once, allow it for the rest of the conversation, or refuse it. Conversations are saved, so you can return to one later or carry an investigation an agent already ran into a chat. The assistant is part of AI agents, which are turned on for each organization individually. See Grepr AI agents.

Manage integrations and datasets with Terraform

You can now create and manage your Datadog, Splunk, New Relic, Sumo Logic, OTLP, and data warehouse integrations, and the datasets inside them, with the Grepr Terraform provider, alongside the pipelines that use them. Reference an integration or dataset from a grepr_pipeline resource and Terraform creates them in order. The new resources need provider version 1.2.0. See Manage integrations and datasets.

Other improvements and fixes

Span searches in the Trace Explorer return several times faster. Benchmarks show certain queries return 5x faster.
You can now collect file statistics on every indexed column of a dataset’s raw log table or raw span table, from Advanced in the dataset’s settings, to make searches faster. See Collect file statistics on every column.
A SQL transform using VARIANT_VALUE now reports an invalid type, behavior, or default when you save the pipeline instead of when the job reads its first log, and a misspelled behavior such as 'DEFALT' is rejected rather than ignored.
An investigation link now opens the transcript over the agent’s Investigations tab instead of on a separate page, so the list keeps its page, filters, and selection while you read a transcript and when you close it. Links you already posted keep working.
The Terraform provider now reports that a pipeline’s tags cannot be changed after creation, and lists the pipeline’s current tags, instead of failing with an invalid plan.
Fixed an issue where a literal open range in a Datadog-like query, such as @code:[* TO *], matched every event instead of testing that the attribute exists, as it does in Datadog.
Fixed an issue where a Datadog-like query that listed many quoted values for one field, such as the link from a reduced log summary to the logs behind it, failed with an INTERNAL_ERROR_QUERY_ENGINE error.
Fixed several issues where an open investigation transcript showed fewer turns than the agent had written, including a missing final turn when a run finished between two refreshes, a stale transcript when you reopened a finished investigation, and stalled scrolling partway down a long transcript.
Fixed an issue where a link to a deleted agent or investigation showed a session error, or sat on a loading spinner, instead of saying that it no longer exists.
Fixed an issue where a VARIANT_VALUE default for a DATE, TIMESTAMP, or TIMESTAMP_LTZ path failed the job on the first log missing that path, and where a numeric default such as 'DEFAULT', 0 failed the pipeline when it was saved.
Fixed an issue where a pipeline whose filter combined a large number of conditions with AND failed to start, and where two comparisons against CURRENT_TIMESTAMP in one query could read different times.
Fixed an issue where an NRQL subquery matched different events than the same condition written at the top level, and where a subquery with no condition of its own matched nothing.

October 6, 2026

Other improvements and fixes

Tag keys on logs sent through a Datadog integration are now converted to lowercase, matching the Datadog convention.
When you add a field to a never-index list, Grepr now removes its column from the dataset’s table on a schedule, 6 hours later by default, which frees the column’s place under the discovery cap. See Stop indexing a field.
Composite aggregation keys on a reducer attribute, such as bytes_sum, now appear on samples, exceptions, and passthrough messages as well as on summaries, so aggregations over those keys count every message.
Fixed an issue where the Trace Explorer ignored the sort order you chose, which made searches over long time ranges slow.
Fixed an issue where the Trace Explorer returned an error for a dataset that had never received trace data.
Fixed an issue where log and span backfills failed with a “Could not submit job” error on some older datasets.
Fixed an issue where a dataset’s table settings returned an error for a table the dataset had not written yet.
Fixed an issue where reducer summaries could contain duplicated or truncated values for an attribute with several composite aggregations that collect lists.
Fixed an issue where a pipeline with a SQL transform whose output goes to the Sinks step failed to start.

October 2, 2026

Faster searches on the log attributes and tags you filter on most

Grepr can now index log attributes and tags as their own columns in a dataset’s data lake table, so a search that filters on an indexed field reads only that column and skips the files that cannot match. In each dataset’s settings, you choose whether Grepr discovers new fields to index, which fields it always indexes, and which it never indexes. Searches return the same results whether or not a field is indexed. See Indexed columns.

Detect log patterns from the shape of their attributes

Log reduction can now build patterns from a log’s attributes instead of only its message, so structured logs whose message is empty or always the same still reduce into distinct, meaningful patterns. On the reducer’s Patterns card, you choose whether patterns come from the message, the attributes, or both, whether attribute values count toward similarity or only the attribute keys do, and which attribute subtrees to include or exclude. These settings replace the single structured tokenization switch, and reduction keeps grouping events by the message alone unless you choose otherwise. See Choose which fields form a pattern.

Import fewer Datadog exceptions and estimate their impact before you build a pipeline

The Datadog integration has a new Active dashboard threshold setting that limits dashboard imports to dashboards viewed within the period you choose, so queries from dashboards nobody looks at no longer produce exceptions.

You can also estimate each imported exception against your indexed Datadog logs, so you can see how much volume it would keep before any Grepr pipeline exists. From the integration’s exceptions dialog, start an estimation run over a scope and time window, and each exception shows the share of indexed logs it matches in the % indexed logs column. Estimating against indexed logs requires an application key with the logs_read_data scope. See Identify log events that should bypass aggregation.

Other improvements and fixes

Log pipelines evaluate filter, exception, and routing queries faster, especially Datadog word searches, case-insensitive matches, LIKE patterns, and long IN lists.
Log pipelines now have a single Parsing step that runs JSON parsing, attribute remapping, and Grok parsing in order, so you can now remap attributes that a Grok pattern extracts.
The Parsed alert queries table now shows when each Datadog dashboard query was last viewed and how many views it has, and you can filter the table on both.
You can now rename an integration, or change settings that no running pipeline reads, such as the active dashboard threshold, without stopping the pipelines that use it.
The Input-Output Lag metric for trace pipelines now includes the time spans wait before the pipeline reads them, so it reports the full delay from ingestion to output. After a pipeline restarts with a backlog, the metric shows the backlog’s real age instead of reading near zero.
The pipeline log viewer now has one tabbed settings dialog in place of its two footer dialogs, and no longer shows a loading spinner in the middle of the results while new logs stream in.
Fixed an issue where Analyze expressions failed with a “Table not found” error when an @ attribute name contained a hyphen or slash, such as @trace-id or @app.kubernetes.io/name.
Fixed an issue where an IN or NOT IN list with a large number of values in an NRQL or Grepr SQL query matched every event instead of filtering.
Fixed several issues in the pipeline log viewer, including column selection, a crash in the popped-out view, the diff tab not working when popped out, and an empty link from the reducer step to the sink step.
Fixed an issue where metrics chart axes showed too few or too many decimal places for small or fractional values, such as CPU usage.