Skip to content

Releases

v0.1.1 is available.

The first public kora release packages the engine, CLI, Helm chart, container images, Python SDK, and MCP server. Artifacts and checksums are published on GitHub Releases at the same time.

Latest release

kora v0.1.1

A self-hosted memory engine for AI agents, with graph and vector retrieval, conversation extraction, project profiles, operational tooling, and a measurable Razorpay receivables agent.

  • REST and gRPC APIs with hashed API-key authentication
  • Hybrid full-text, vector, and graph retrieval
  • Helm chart, backups, metrics, and production probes
  • CLI binaries, Python SDK artifact, and MCP server
  • Docker images for the API, web UI, and PostgreSQL stack
  • Agent integration and Razorpay example documentation

This pre-1.0 release includes the Context0 to Kora rename. Existing deployments must follow the migration notes before upgrading.

Changelog

What changed in v0.1.1

Breaking: Context0 is now Kora

Environment variables moved from CONTEXT0_* to KORA_*, the Go module and gRPC package were renamed, and the default PostgreSQL role and database are now kora. The AGE graph schema remains context0 so an upgrade does not require a graph migration.

Existing deployments should run the rename migration script or keep the previous PostgreSQL user and database through Helm values.

Agent integrations

A seven-tool MCP server, Python SDK, REST API guide, and the measured Razorpay receivables agent.

Retrieval quality

Full-text, vector, and graph signals are fused per query, with superseded memories demoted from current answers.

Production operations

A hardened Helm chart, backup and restore, structured logs, Prometheus metrics, and split health probes.

Security

Hashed API keys, deny-by-default authorization, secret-free chart defaults, and safer web UI credential handling.

Selected fixes

  • Hybrid retrieval now changes result order instead of only reporting scores.
  • Scoped vector queries return the requested result window without silently dropping matches.
  • Write-time folding stops repeated conversations from filling the graph with restatements.
  • Invalid consolidation settings fail validation instead of deleting memories with fallback values.
  • Session accounting, database outage behavior, and REST rate limiting are covered end to end.

How versions will work

Semantic versioning, with the pre-1.0 caveat that matters: while the major version is zero, a minor bump may contain a breaking change. Pin an exact version until 1.0, and read the release notes before upgrading.

Patch
Fixes and hardening. Review upgrade notes before 1.0.
Minor
New capability. Before 1.0, may break something.
Major
Reserved for 1.0 and the stability promise that comes with it.