Release Notes - v1.1.1

Release Date: 2026-05-29
Changes: v1.1.1.dev0 → main

Pull Requests: #2658

Summary

Cognee 1.1.1 focuses on improving agents, visualization, and graph handling: agent lifecycle and connection management are reorganized and exposed via clearer endpoints and SDK support; the visualization UI and Story layout were rebuilt for clearer graph storytelling; and graph ingestion and Neo4j/Postgres handlers gained robustness and new features. Several bug fixes and CI/test improvements increase stability and developer experience.

Highlights

• Revamped agent management: new endpoints, connection naming, unregister/deactivate flows, and SDK access to agent configurations.
• Major visualization rewrite: pipeline-aware Story layout, schema view, clearer legend and UI controls, and better rendering of edges/labels.
• New/updated graph dataset handlers (local Neo4j support) and improved edge ingestion and fallback labeling.
• POST /v1/remember gains a graph_model form parameter to influence remember behavior.
• Multiple bug fixes for prune, cognify/remember routing, and LLM/API handling to prevent silent failures.

Breaking Changes

• ID types changed: internal IDs were converted to UUIDs in places where they previously used plain strings. Clients or integrations that rely on exact ID types or formats may need to be updated to handle UUIDs.
• Agent endpoint changes and renames: agent-related endpoints were split and renamed to better distinguish between agent resources and agent connections. If you have custom tooling or integrations that call agent endpoints, you must review and update calls to the new route names and parameters (including new/session name requirements on some endpoints).
• Dataset parameter rename: dataset_name was renamed to dataset in API/CLI places — update scripts/automations that still use dataset_name.
• Environment variable rename: some local Neo4j-related env var names were simplified (e.g., from local to neo4j). Check deployment configs.
• Remember routing change: remember(session_id=...) is now routed to the JSON /entry endpoint. If you relied on prior implicit routing behavior, verify your flows.

New Features

• Agent management: added new agent endpoints and split existing agent APIs so agent lifecycle and connections are easier to manage. Agents can now define unique names for their connections and unregistering marks agents as inactive.
• SDK: added ability to fetch a specific agent configuration directly from the SDK (useful for programmatic inspection and tooling).
• Visualization: introduced a rebuilt Story layout and Schema view. New visual cues surface importance, feedback and ontology hints; bottom legend and simplified controls improve readability and exploration.
• Graph datasets: added a local Neo4j dataset handler and expanded support for POST /v1/remember with a graph_model form parameter (for better graph memory control during recalls).
• Global Context Index: new bucketing and placement strategies, scoring logic, and build/update flows to improve global context extraction and retrieval.

Improvements

• Agent connections: simplified connection identifiers (agents may provide a unique name), clearer separation between agent resources and their connections, and a new endpoint to check current user connections.
• Edge ingestion: improved fallback label generation (from metadata.index_fields) and corrected label rendering so relationship labels appear inside --[ ]--> markup rather than edge_text.
• Resilience: edge processing now skips edges with unprojectable endpoints instead of raising an exception, reducing failures during ingestion and visualization.
• Databases & ingestion: better error handling for dataset lookups and prune operations; Postgres/pgvector helpers and admin scripts received upgrades to ease setup and maintenance.
• CI / developer experience: many workflow updates (including optional LLM_ARGS) and added tests to CI to catch regressions earlier and increase cross-DB coverage.

Bug Fixes

• Fixed prune-related errors on the handler side and added protections to catch prune exceptions so background maintenance is more reliable.
• Cognify / remember flow fixes: removed DatasetConfiguration DB lookup/write-back that could cause issues; routed remember(session_id=...) to the JSON /entry endpoint to avoid silent no-ops; added a ValueError when custom_prompt is used with session_id in API mode to avoid unexpected behavior.
• LLM integrations: honor OpenAI instructor mode and fixed related LLM client handling.
• Data handling: soft-fallback to type name when a DataPoint has no labelable field to avoid missing labels, and fixes to batch query top_k tuning for brittle tests.
• Visualization fixes: fixed node spacing and pinned X position in Story mode to avoid column violations and improved edge skipping when endpoints can’t be projected.

Technical Changes

• Large visualization refactor: replaced older single-file visualization implementation with a modular preprocessor and JS-driven views (Story, Schema) and templates for better maintainability and UX.
• Refactored agents: moved agent mode into modules/agents, introduced agent registry and operations modules, and separated agent models. Added persistence for current agent configuration.
• Graph & index internals: new placement and scoring modules for the global context index, moved and renamed many memify/global_context_index modules for clearer responsibilities.
• Database adapters: added/rewrote dataset handlers for Neo4j and improved Postgres/pgvector support and table creation flows.
• Tests & CI: added numerous tests (unit, integration, e2e), adjusted workflows to accept optional LLM_ARGS, and reduced brittle/flaky tests by simplifying or replacing fragile checks.

Generated by Cognee Release Notes Generator on 2026-05-29