From fef1ab28e21a83785db7bdd125757a6f01b7138d Mon Sep 17 00:00:00 2001
From: jedigalt <201954051+jedigalt@users.noreply.github.com>
Date: Mon, 9 Nov 2026 12:21:17 +0000
Subject: [PATCH 1/7] [needs-review] docs(release-notes): week of 2026-11-02
---
src/pages/docs/release-notes.mdx | 56 ++++++++++++++++++++++++++++++++
1 file changed, 56 insertions(+)
diff --git a/src/pages/docs/release-notes.mdx b/src/pages/docs/release-notes.mdx
index ca9500c1e..c888b9e3a 100644
--- a/src/pages/docs/release-notes.mdx
+++ b/src/pages/docs/release-notes.mdx
@@ -5,6 +5,62 @@ description: "Latest Future AGI release notes covering new features, improvement
{/* release-notes:insert-below — automation inserts new releases here; do not remove */}
+## Week of 2026-11-02
+
+
+
+
Features
+
+- **Slack Alerts for Continuous Eval Tasks:** A continuous evaluation task now posts a Slack alert when it fails or stalls, so a silently stuck task no longer goes unnoticed.
+
+- **Open the Source Trace From Eval Task Usage:** From any run on an eval task's Usage tab, Open trace now opens the exact trace with the evaluated span selected in a drawer, without leaving Usage, and previous/next steps through the task's traces in order. A voice project opens the voice call drawer instead.
+
+
Bugs/Improvements
+
+- **Environment Workspace Fixes:** Hover tooltips were removed from the environment workspace tabs, and an environment's scenario count now updates live instead of freezing at the first value it loaded. The environment builder chat also now offers to approve every change in a request at once, instead of confirming each edit separately.
+
+- **Faster, More Reliable Environment Builds:** An agent that reads its assigned network port now builds and tests across several workers at once instead of one at a time, and an agent backed by its own embedded SQLite database now builds correctly instead of failing. Environments built on an earlier platform version also keep working after an upgrade, and a build or check run under heavy load no longer fails from a lost results file.
+
+- **Debug Failures Groups Duplicate Findings:** Debug failures investigations now group accepted findings together instead of listing the same failure more than once, and revisit findings that were set aside instead of dropping them. A goal with no findings no longer shows as an unexplained row.
+
+- **Error Feed Issues Reflect Only Live Eval Results:** An Error Feed issue no longer keeps listing a trace after its evaluation result has been deleted, and an issue's score, turns, and Trends statistics now reflect only live results instead of reading zero when nothing remains to measure.
+
+- **Eval Task List Shows Accurate Last-Run Times:** An eval task in the Tasks list now shows when it last actually produced work, instead of continuing to show its creation date no matter how many times it had run since.
+
+- **More Reliable Few-Shot Evaluation Inputs:** Few-shot evaluation inputs that are zero or false are no longer dropped as if they were empty, and structured (JSON) inputs are now matched and sent to the embedding model correctly instead of failing outright.
+
+- **Prevent Silent Blank Dataset Rows:** Writing dataset rows through the platform's AI assistant or its MCP tools with an unrecognized column, or in a malformed batch, is now rejected with a clear error instead of silently saved as blank rows.
+
+- **Gateway Streaming and Cost Accuracy:** A request that times out or hits a provider error mid-stream now ends with a proper error event instead of cutting off silently, across chat, Messages, Responses, Completions, GenAI, speech, and Assistants run streams. Streamed Responses calls, plus image, rerank, search, OCR, and embedding requests, are now priced or clearly flagged as unpriced instead of appearing free, a stream cut off before usage arrives is now priced from an estimate instead of showing no cost at all, and pricing catalogue gaps are now filled from additional public sources. Cost and latency metadata now reach every streaming route instead of chat alone, and Messages requests now report the timeout that was actually applied.
+
+- **Gateway Billing Now Covers Every Request Type:** Assistants runs and Vector Store calls now go through the gateway's usual budget and request-logging checks instead of running unbilled and unlogged. Several provider pricing lookups are also fixed so models offered under more than one provider name, or through a reseller, are billed at their real price instead of silently at zero.
+
+- **Gateway Budgets, Guardrails, and Alerts on Every Route:** Budgets now price spend from the live model catalogue instead of an outdated fallback table, input guardrails see request text on the native Messages and Responses routes, a blocked response returns the correct error on every route including streaming, and budget alerts are now actually sent.
+
+- **Gateway Request Log and Provider Fixes:** A request is now logged exactly once instead of several times, gpt-5.6 models support tool calls over the Responses API, a malformed metadata header is now rejected instead of silently dropped, and an organization can add Perplexity and more than one custom provider. Gemini tool calls also get a unique id per call instead of ids that repeat across responses, which previously broke agents that track calls by id across multiple rounds.
+
+- **Gateway Model List Matches Key Access:** The gateway's model list no longer advertises provider models an API key is not entitled to call; it now lists only the models the key can actually use.
+
+- **Guardrail Credentials and Request Logs Stay Scoped to Their Organization:** A gateway configuration could previously decrypt another organization's stored guardrail credentials, and gateway traffic spanning multiple organizations in one batch could file requests, webhook events, and usage under the wrong one. Both are now resolved per organization before anything is stored or decrypted.
+
+- **More Reliable Guardrail and Gateway Config Saves:** Activating a Guardrails or Fallbacks change no longer resets an organization's other settings, such as budgets, usage limits, and IP restrictions, back to their defaults, and toggling or editing a guardrail no longer fails for organizations managed through the Settings editor. The UI also now warns clearly when a save or rollback did not reach the gateway, instead of reporting success.
+
+- **Guardrail Testing Shows the Real Outcome:** The Guardrails Test tab now shows a warning for rules in Warn or Log mode instead of reporting a plain OK, along with the real HTTP status code. A guardrail whose authentication fails is now treated as a failure instead of silently calling the check with no credentials.
+
+- **Future AGI Eval Guardrail Works on Self-Hosted Installs:** The Future AGI Eval guardrail no longer defaults its Base URL to our cloud endpoint, so self-hosted organizations get their own credentials injected instead of the check running with no, or the wrong, credentials.
+
+- **Trace Ingestion Limits:** The trace collector now enforces per-request limits on OTLP payload size, element count, and nesting depth, and authenticates gRPC connections before decoding them, so an oversized or malformed request now gets a clear error instead of exhausting memory.
+
+- **SQS Export Reliability:** The SQS export integration now retries individual failed messages and only advances past a batch once every record in it is confirmed delivered, instead of silently skipping undelivered records.
+
+- **Annotation Labels Filter Order:** The Labels filter toolbar now matches the Queues toolbar's order (search, then Active/Archived, then type), for consistent navigation between the two.
+
+- **Self-Hosted Windows Startup Fix:** Self-hosted deployments checked out on Windows now start correctly; a line-ending mismatch in shell scripts and container build files previously stopped the stack from starting.
+
+
+
+{/* SKIPPED: internal-only churn (version-bump and branch-sync merges, generated-contract and tool-manifest regeneration, test-only changes); two local-development-only contract-validation fixes confirmed to have no effect in production; a short-lived SDK issue introduced and fixed within this same window with no net customer impact; an internal operational incident whose only customer-facing outcome is the Slack-alert item above. Held for confirmation: environment, agent and scenario-suite versioning with run comparison and rollback (merged but still behind platform flags that were off in production as of this release); a capability for repairing a failing agent from a failed simulation run (its screens shipped, but the backend run pipeline still depends on a production template update that was not yet complete); multi-agent framework reference fixtures (partial hosted coverage per the author); runtime sync of the pricing catalogue from a public source (only the static half merged); nesting of sub-agent traces under their parent call (the fix lives in a dependency outside this week's reviewed repositories); comparing an agent's turns against its own prompt and tool calls as part of a Debug Failures investigation (the data is now sent, but the comparison logic itself lives in a dependency outside this week's reviewed repositories, so it cannot be confirmed as live). */}
+
## Week of 2026-10-19