v1.5.0 — Large-scale migration & graph reliability
Release Date: 2026-08-15
Changes: v1.5.0.dev2 → main
Pull Requests: #4477, #4504, #4505
Summary
This release focuses on making large-scale dataset migrations reliable and faster, with many fixes and performance improvements to the Ladybug graph adapter and the migration pipeline. It also adds community plugin documentation, better LLM tuning options (temperature and seed), and CI/performance tests that validate production-shaped migrations at scale.
Highlights
- Large-scale migration validation: new release-test CI that seeds a production-shaped two-dataset system and runs the full migration chain to verify fork-splitting and rekey operations.
- Ladybug (graph adapter) stability and speedups: set-based fork re-key, chunked queries, batched rekey/restore, and fixes that make graph migrations more reliable and faster.
- Community plugins and migration docs: new cognee-community skill documentation and a tutorial for migrating from Letta/MemGPT and Zep to Cognee.
- LLM tuning and reproducibility: plumbed LLM_TEMPERATURE and added LLM_SEED support for reproducible outputs when the provider supports seeds.
- Windows support and observability hardening: ship OpenSSL for Windows to avoid TLS issues with Ladybug and batch OTLP log export while honoring tracing-off switches.
Breaking Changes
- No user-facing breaking changes in this release. Migration behavior was hardened and tested more aggressively, but there are no required changes to existing user configuration or APIs.
New Features
- Large-scale migration release test: A CI workflow seeds a production-shaped mock system (two datasets) using the pinned legacy Cognee, runs the complete migration chain on it, and verifies dataset-scoping splits and rekey_fork_document_ids. What this does: automates heavy migration validation so releases are verified against production-like data. Why it matters: catches data-migration problems before they reach users and reduces risk for upgrades.
- cognee-community documentation (new skill): A new guide that explains how to use and contribute community-maintained plugins (database adapters, connectors, and tasks). What this does: shows how to install, register, and use community packages such as Qdrant, Milvus, Slack/Gmail connectors, and more. Why it matters: makes it easier to extend Cognee with third-party adapters and connectors while avoiding common configuration pitfalls.
- Migration tutorial from Letta/MemGPT and Zep: A step-by-step migration guide and example dataset dumps to help move data into Cognee. What this does: provides concrete instructions and sample files to migrate from popular alternative systems. Why it matters: lowers the barrier to move existing memories into Cognee without manual export/import guesswork.
- LLM temperature and seed controls: Support for an LLM_TEMPERATURE environment setting (controls sampling temperature) and a new LLM_SEED setting (sampling seed for reproducible outputs when supported by the provider). What this does: sends temperature and optional seed with every LLM call when set. Why it matters: gives users control over randomness and reproducibility of LLM outputs across runs.
Improvements
- Set-based fork re-key on Ladybug: Reworked the re-key algorithm to operate on sets rather than single-item operations. What this does: reduces the number of database operations during dataset fork re-keying. Why it matters: significantly improves speed and reliability when splitting or rekeying large graphs.
- Batched migration re-embed and rekey restore calls: Vector re-embed and rekey restore work in batches instead of per-item calls. What this does: reduces load on embedding services and the database during migrations. Why it matters: lowers migration time and external API pressure for large datasets.
- Chunked edge-identity queries and write chunking tests: Queries that enumerate edge identity are now chunked and tests strengthened for write-chunking behavior. What this does: avoids oversized queries and makes large write operations safer. Why it matters: prevents timeouts and failures when performing large graph writes or rekeys.
- Trace and logging improvements: Batch OTLP log export and honor the tracing-off switch so tracing and log exports respect configuration. What this does: reduces tracing/logging overhead and prevents spurious network usage when tracing is disabled. Why it matters: improves observability performance and respects user configuration.
- CI and performance harness: Added a nightly 100k-node/-edge 'War and Peace' large-mock performance benchmark and adjusted release-test sizing and permutations. What this does: runs a production-scale stress benchmark in CI and matrices migration tests across adapter pairs. Why it matters: catches scale regressions early and validates migrations across different backend combinations.
Performance
- Faster rekeying and migration: Set-based fork re-key and batched re-embed calls reduce total migration time on large graphs (practical impact: lower wall-clock migration duration and fewer database round-trips).
- Reduced query pressure: Chunking edge-identity queries and batching rekey/restore operations reduce peak DB load and likelihood of timeouts during large operations.
- Nightly large-scale benchmark: The new 100k-node/edge mock benchmark surfaces performance regressions on production-shaped workloads so real regressions are detected earlier.
Security
- Windows OpenSSL bundled for Ladybug: The release supplies OpenSSL for Windows to avoid TLS/SSL failures when Ladybug is used on Windows. What this does: ensures secure connections work reliably on Windows hosts running the Ladybug adapter. Why it matters: avoids connectivity failures and potential insecure fallbacks on Windows deployments.
Bug Fixes
- Bind edge endpoints in a single MATCH clause: Fixed an issue in graph queries where edge endpoints were bound in separate clauses, which could cause incorrect matches or inefficiencies. Result: more reliable and correct graph traversals during migration and queries.
- Pin and align Ladybug versions: Adjusted pinned Ladybug versions and aligned minimum floors across branches to avoid compatibility mismatches during migration tests.
- Chunk ladybug edge-identity queries and batch rekey restore: Prevent oversized queries and failures during restore operations by chunking and batching.
- Settle first batch before concurrent upserts: Fix to reduce race conditions when doing concurrent upserts during bulk operations.
- Batch OTLP log export and honor tracing off switch: Fix ensures logs are exported in batches and that disabling tracing stops exports as expected.
- Fixes to various migration tests and release-test matrix: Multiple test fixes to ensure migration compatibility runs reliably in CI.
Technical Changes
- CI/workflow updates: Expanded release-test workflow to include nested dev canary release permissions and a new large-scale migration compatibility job; added nightly performance workflow with a 100k-node mock scenario.
- New test and tooling files: Added many migration and mock-ingestion test utilities to exercise large-scale scenarios and validate batching behavior.
- New helper for Windows OpenSSL: Added cognee_db_workers/_windows_openssl.py to distribute OpenSSL for Windows users running Ladybug.
- Refactors and housekeeping: Minor code cleanups, comment removal, and version bump to v1.5.0; multiple fixes across graph adapter and migration modules to support batched operations and chunked queries.
Dependency Updates
Updated:
- ladybug: >=0.16.0,<=0.18.2 → ==0.19.0
Compatibility
| Component | Supported / Required |
|---|---|
| Python | >=3.10,<3.15 |
| pydantic | >=2.10.5 |
| litellm | >=1.83.7 |
| fastapi | >=0.116.2,<1.0.0 |
| sqlalchemy | >=2.0.39,<3.0.0 |
| lancedb | >=0.24.3,<1.0.0 |
| ladybug | ==0.19.0 |
— The Cognee Team · 2026-08-15