This release delivers three customer-visible capabilities: Cognis Memory, Initiative Discovery, and Atomic Code Linking, alongside sprint lifecycle automation, MFA self-service, and a stack of reliability fixes.

Every AI-first founder is moving at a pace where the strategy honestly evolves between Monday and Friday. Which creates a real problem: how do you tell the business what you’re building when “what you’re building” is happening at a rate no one anticipated?

This release is the foundation for what we’re calling Autonomous Intelligence. A system that reads execution, derives strategy, and communicates it back to the business without anyone maintaining metadata.

Three capabilities compose into it. Initiative Discovery reads your commits and tells you what your initiatives actually are, derived from the work, not declared in a slide deck. Cognis Memory gives the reasoning engine three layers of persistent context across every thread, so the model sharpens every sprint. Atomic Code Linking ensures every commit gets automatically linked to the right initiative, even without ownership rules.

What this composition enables: ask Cognis “why is this sprint slipping?” and it returns the missed commitment, the owner, the slot that was never scheduled, the cross-repo blocker. Every claim cited by commit SHA. That’s not generic AI. That’s an engineering intelligence layer that reads your execution, remembers your context, and reasons over both.

For private equity, or any CTO or VP Engineering taking over a new team, the same composition unlocks a new use case: an operating partner or incoming leader walks into a new codebase, connects InteliG, and within minutes sees what the engineering team is actually building. Not what the deck says. Not what the founder claims. What the commits prove.

Cognis Memory

Cognis now has memory. Three layers deep, persistent across every thread and every session.

User Memory. Per-user durable facts that survive sign-out. Your thesis, your role, your operating cadence, what you’ve told Cognis to watch. Visible as the “You” section of the right rail in every thread.
Org Memory. Per-organization context attached to every Cognis response. Company, top priority, team size, tech stack. Visible as the “Org” section of the right rail.
Thread Memory. A tool-callable scratchpad the LLM uses to carry facts forward inside a single conversation, without re-reading the transcript every turn.
LLM reflection job. A nightly process reads your real activity and writes new durable facts to User and Org memory. Auto-extracted entries are labeled “inferred” so you can see what came from you versus what Cognis observed.
Provenance and bitemporal storage. Every memory entry knows where it came from, when it was written, and when it became valid. You can trace any Cognis answer back to the memory it was grounded in.
Conflict-aware writes. When new information contradicts existing memory, Cognis merges and keeps history rather than silently overwriting. The prior value remains visible.
Inspectable and editable. A dedicated settings page lists every memory entry. Inspect, edit, or delete any of it. Memory is yours, not Cognis’s.
Quota and size caps. Memory doesn’t grow unbounded. Old low-signal entries age out automatically.

Initiative Discovery

Propose a structured strategy — themes, initiatives, and workstreams — directly from your repository activity. Onboard with an existing codebase and InteliG returns a working draft of how the work organizes, grounded in commits.

Universal signal enrichment. Branch names, file paths, and conventional commit prefixes feed the proposal pipeline, so discovery produces meaningful results before any initiative is declared.
Window control. A date-range picker lets you run discovery over the last 14, 30, 60, or 90 days. Useful for organizations with deeper commit history.
Supporting commits are visible. Each candidate shows the specific commits that informed the grouping, including branch names and file paths. You can verify why every proposal exists.
Duplicate detection. Discovery identifies when a proposed initiative overlaps with work you’ve already declared, surfaces the overlap percentage, and lets you merge or keep separately.
Evidence-driven grouping. Existing initiatives are matched via commit evidence, not name similarity, so new work surfaces as distinct proposals rather than collapsing into existing names.
Workstreams are evidence. Confirming or rejecting an initiative automatically cascades to its workstreams. Workstreams are inspectable in the detail view, and the pending count reflects only initiative-level candidates.
Confirm creates a real initiative. Accepting a discovery candidate immediately creates an active initiative under the correct quarter and theme, attributed to the supporting commits.

Atomic Code Linking

If a commit doesn’t match an explicit ownership rule, Cognis links it to the right initiative automatically. Every commit becomes its own atomic linked unit. No gaps, no manual setup.

A single AI classifier replaces the heuristic chain. A 9-strategy matcher (keyword, topic, title, phrase, classification, semantic, LLM, aggregator) produced flat-cluster noise on homogeneous codebases — every commit ranked equally similar to every initiative. About 1,500 lines deleted. In its place: an enum-constrained classifier that returns one proposal per commit at the model’s actual confidence, with a calibrated rubric and a no-match cache so repeated declines don’t pay twice.
Ownership patterns are unchanged. Commits matched by file globs or explicit initiative IDs still auto-link as before. The classifier is the fallback for everything else.
Alignment scores read from real evidence. Verified alignment percentage reads from the commit-level foreign key and match type, so the score reflects actual links.
Pull requests are first-class signals. A new pull-request signal indexing agent treats PRs as initiative-intent signals alongside commits.
One page, every pending decision. A new org-wide Initiative Code Linking Review surface consolidates all pending AI suggestions and confirmed links in one place. Filterable by initiative, match type, confidence band, and free-text search across commit titles and narratives.
Manual attribution override. On any commit or pull request, you can override InteliG’s attribution with an explicit pick and a reason. The change is recorded in a full audit trail visible on the detail view, and supersedes any pending AI proposal for the same row.
Bulk actions. Select multiple suggestions and confirm, dismiss, or reassign them in one action. A “Confirm all high-confidence” shortcut surfaces when high-confidence suggestions are present.
Alignment lift indicator. The review surface shows projected alignment-score improvement as you confirm suggestions. Reviewers see the concrete impact of each decision before committing to it.
Commit narrative inline. Every AI proposal row renders the commit narrative and short SHA, so reviewers don’t have to context-switch to the commit detail view for routine decisions.

Cognis

Beyond memory, Cognis got a major capability lift across the surface: release artifacts generated inline, strategy artifacts drafted from chat, reasoning over its own agent fleet, and a redesigned right rail that surfaces real organizational context with every answer.

Release notes generated in one turn. Asking Cognis to generate a release completes inline and returns a direct link to the artifact. No separate approval step. The artifact stays in Draft until you review and publish.
Cognis proposes strategy artifacts in chat. New mutating tools let Cognis draft a Vision, Theme, Initiative, or Roadmap directly from a conversation. Each draft flows through the same human-review surface as any other proposal — Cognis suggests, you confirm.
Cognis reasons over its own agent fleet. Cognis can answer questions about which agents are registered, when they last ran, and what they found — backed by a 33-tool catalog spanning agents, commits, contributors, findings, action items, and transcripts. No more hallucinated agent names.
Right-rail Execution section. A new section surfaces declared-versus-inferred sprint commitments side by side. See at a glance where the team’s actual execution diverges from what was committed, with one-click drill-down on every commitment.
Redesigned right-rail context panel. The right rail organizes the thread into You, Your org, Execution, Findings, and Reasoning sections so every answer is grounded in the same context Cognis is seeing.
Proactive findings, grouped and continuous. The Findings panel rolls up related findings (for example, bus-factor risk across multiple repositories for the same contributor) into a single row with drill-down. A new dual-writer keeps the findings feed warm continuously, not just when someone asks.
Live agent-loop pulse. The Reasoning rail shows Cognis’s tool calls in real time as a thread streams, so you can see what tools it’s reaching for, in what order, and what they return.
15 registered agents, 4 categories. Cognis Reliability, Agent Intelligence Platform, Strategy, and Identity & Security. The full fleet is visible on a single page with category filters and live run-status indicators.

Sprint management

Sprint autopilot toggle. A toggle on the Sprints page lets you opt your organization into automatic end-of-sprint hygiene — creating the next sprint, inheriting active initiatives, and rolling open action items forward — without requiring a configuration change.
Sprint names. Sprints now have human-readable names (defaulting to “Sprint 1”, “Sprint 2”, etc.) that appear as the primary label everywhere a sprint is referenced. Names are editable inline on the sprint detail page.
Action-item sprint membership. Action items can be pinned to a specific sprint, moved between sprints, or unpinned directly from the Knowledge Action Items page and the Sprint detail tab.
Three-section Sprint detail. The Sprint detail tab now organizes action items into Pinned (committed to this sprint), Inherited (rolled from a prior sprint), and Suggested (meeting-derived, within the sprint window).
Strategy navigation reorganized. The Strategy sidebar is grouped into Direction, Alignment, and Delivery sections, making it easier to find the right page without reading every label.

Security and access control

Self-service MFA disable. Users can remove their authenticator app from Security Settings using either a 6-digit TOTP code or a recovery code. When your organization requires MFA, the removal option is disabled with explanatory copy.
Authentication path hardening. Internal authentication routes are gated behind explicit environment-level enablement, with audit logging on every invocation and a locked default posture across all environments.

Reliability fixes

GitHub repository uniqueness enforced. A race condition that created duplicate repository records during initial sync — causing errors when pushing to GitHub — is now prevented at the database level. Existing duplicates are resolved on upgrade.
Initiative lifecycle agent is window-aware. The agent identifies cross-period attribution drift accurately, preserving historical commit links while clearing only genuinely misattributed ones.
Session refresh honors active org. Session refresh respects the user’s currently active organization rather than reselecting the primary org.

Why it matters

The full attribution loop is now end-to-end. From LLM-powered proposals to org-wide review to alignment-lift measurement, reviewers can drive alignment-score progress without commit-by-commit manual attribution. The engine underneath produces calibrated, confidence-scored proposals instead of heuristic noise.
Cognis is now a working surface, not just a Q&A box. It generates release notes inline, drafts Visions, Themes, Initiatives, and Roadmaps from conversation, and reasons over its own agent fleet with a 33-tool catalog. The redesigned right rail makes its work legible — what it’s reasoning about, what it knows about your organization, and what it’s found.
Sprint lifecycle automation removes recurring end-of-sprint coordination overhead for teams that opt in. Sprint names, action-item pinning, and the three-section detail view make the sprint surface usable as a daily working view.
MFA self-service removes a support-ticket category without weakening enforcement when your organization requires MFA, while authentication routes are gated and audited by default across environments.

Release confidence

Scope. intelig-backend (main), intelig-frontend (main), window 2026-05-13 to 2026-05-20.
Evidence. 200+ commits with detailed descriptions across both repositories. Five merged pull requests (#78 Cognis memory, #79 MFA self-disable backend, #27 Cognis memory UI, #28 Cognis context panel, #29 MFA self-disable frontend). Several features confirmed with live end-to-end test results cited in commit descriptions: discovery runs verified against multiple real orgs, MFA disable covered by integration test, repository uniqueness fix verified on live duplicates.
Confidence. High for Cognis Memory, Initiative Discovery, Atomic Code Linking, MFA self-service, security hardening, and the repository uniqueness fix — all include explicit verification evidence and replaced or repaired code paths with measurable behavior changes. Medium for the Cognis surface improvements (release generation, strategy proposals, agent-fleet reasoning), which are verified by integration tests and live usage but represent new surface area. Sprint lifecycle automation is gated behind an opt-in toggle, limiting blast radius until customers turn it on.

Cognis Memory, Initiative Discovery, and Atomic Code Linking