Release Synthesis, Standing Agents, and a Stronger Cognis

This release delivers three customer-visible capabilities: Release Synthesis, a complete Microsoft Teams meeting integration, and a new suite of standing automation agents, alongside a major rewrite of the Cognis reasoning pipeline and meaningful security and reliability improvements across the platform.

Release Synthesis

Generate customer-ready release notes, executive briefs, and technical changelogs directly from your code, your source of truth.

• One artifact, multiple formats. From the new Releases workspace, pick a date range, repositories, and an artifact type (Customer Notes, Executive Brief, Technical Changelog). Cognis reads every merged PR and commit in the window and produces the artifact in seconds.
• The artifact is structured. Each generated release is backed by an explicit evidence trail: PR count, commit count, repositories, and source records, all linked back to git.
• This page is the receipt. The notes you are reading were generated by Release Synthesis from the May 12 window. No human typed the summary.

Microsoft Teams meeting integration

Teams is now fully connected end-to-end.

• After completing OAuth admin consent, meeting transcripts are automatically ingested in real time via Microsoft Graph subscriptions. No manual action is required after setup.
• Historical meeting import lets you backfill past meetings from a date range, accessible from the Teams integration card.
• Improved setup guidance. The setup wizard now walks admins inline through the required Microsoft Application Access Policy (PowerShell steps) and the Test Connection check explicitly tells you whether meetings are accessible or blocked, with the fix steps shown immediately if blocked.

Standing automation agents

Ten standing agents now run on top of the Cognis reasoning engine, keeping the data clean so the reasoning stays trustworthy.

• Execution Catch-Up repairs historical gaps in engineering data after onboarding, with configurable date ranges, dry-run preview, and apply controls.
• Action Item Closure automatically closes resolved meeting action items based on commit evidence. When a linked commit merges, the action item closes itself.
• Commit Narrative generates plain-language descriptions of commits on demand from the commit detail view, and refreshes the commit’s quality score using the narrative as additional context.
• Release Artifact Synthesis (the agent behind this page) generates customer-facing release notes, executive briefs, and technical changelogs from merged pull request and commit activity across a date range.
• CTO Briefing generates executive briefings for engineering leaders from meetings, action items, commits, initiatives, sprint health, and peer-agent findings.
• Agents now display clear readiness and blocking indicators. When an agent cannot run, the reason is shown rather than silently failing. Errored agents can be manually retried.
• Agent category filters are now available on the agents page, and the filter bar scrolls horizontally on narrow viewports rather than overflowing.

Cognis reasoning engine

The Cognis reasoning pipeline was substantially rewritten to make answers more accurate, more grounded, and more auditable.

• Tool catalog and dispatcher. Queries are now dispatched through an explicit tool catalog with dedicated tools per question type (code, knowledge, docs, identity), replacing the prior keyword-based routing.
• ReAct reasoning loop. Cognis decides which tools to call based on the question, rather than running a fixed sequence. Answers cite better evidence and surface less irrelevant context.
• Shadow validation. Every Cognis pipeline change now runs against the previous behavior in shadow mode, so reasoning improvements ship without regressions.
• Per-request audit telemetry. Every Cognis response is auditable end-to-end. Tripwires detect identity-bundle dumps and other unsafe outputs before they reach the user.
• Faster multi-turn responses. Cognis no longer reloads the full identity bundle on follow-up questions, reducing response latency on longer threads.

Security and access control

• MFA hardening. Multi-factor authentication challenge lookup is now bound to the org context derived from the challenge token itself, closing a gap where the pre-authentication lookup ran outside the tenant isolation boundary.
• MFA secret versioning is now supported, allowing key rotation without breaking active sessions.
• OAuth token encryption is enforced for stored Teams credentials.

Reliability fixes

• Fixed a bug that caused GitHub webhook events from personal (User-type) GitHub App installs to be silently dropped. All commits and pull requests from those accounts now process correctly. Existing unprocessed events are recovered automatically on the next retry cycle.
• Fixed the “Manage repository access on GitHub” link for personal GitHub accounts, which previously 404’d and prevented users from expanding repository access.
• Cognis scope filtering now works correctly: selecting a specific contributor or team narrows results as expected, and team-membership questions (“Who is in the Product team?”) route to the right data source.
• Hidden contributors no longer appear in the Cognis scope dropdown.
• The contributor scope dropdown now scrolls when the list exceeds the viewport.
• The Cognis chat interface is hardened against stalled streams that previously left the loading indicator stuck indefinitely.
• AI-generated commit quality rationales are no longer silently truncated.
• Knowledge integration artifact counts and webhook activity filtering are now accurate.

Why it matters

• Engineering leaders can now generate a customer-ready release note from a date range in seconds, instead of spending half a day stitching together a draft from PR titles.
• Teams customers can connect once and receive ongoing transcript intelligence automatically, with clear setup guidance that eliminates the silent “0 meetings imported” onboarding failure.
• Engineering teams gain automated agents that reduce manual follow-up work: closing stale action items, filling historical data gaps, and generating commit context without manual documentation.
• Security and compliance posture is strengthened through tighter tenant isolation at the MFA layer and encrypted credential storage for OAuth integrations.
• GitHub users on personal accounts who were previously receiving no value due to a silent data-ingestion bug are now fully supported.

Release confidence

• Scope: Two repositories (intelig-backend, intelig-frontend). Most of this work landed in the five days from Friday, May 8 to Tuesday, May 12.
• Evidence: 100+ commits and 10 merged pull requests with explicit descriptions, test results, and validation notes.
• Confidence: Teams integration, agent capabilities, the Cognis Harness rework, security hardening, and the GitHub User-account fix are all well-evidenced with tests cited. The release synthesis feature itself is newly shipped and actively used to generate this document. Operational validation of Teams real-time subscriptions in production depends on tenant-specific Microsoft Graph policy configuration by the customer’s admin.