This is a major memory release. Frona introduces an ontology-backed Personal Knowledge Management (PKM) system that turns conversations, facts, entities, relationships, and successful workflows into a structured and inspectable long-term memory. Background consolidation grounds extracted knowledge in source material, resolves identities, evolves each user’s ontology, and publishes readable Markdown pages that agents can search and cite.
The release also substantially refactors inference, chat history, channel supervision, model-provider configuration, sandboxing, and container development. It includes broad reliability, performance, testing, dependency, and user-interface improvements.
Existing installations remain on the basic memory backend unless PKM is explicitly enabled. Fresh installations select PKM during setup. I highly recommend creating a backup before upgrading.
Ontology memory performs background extraction, classification, reconciliation, and page authoring, so the best model is not necessarily the largest or most expensive one. We recommend DeepSeek V4 Flash, Luna with medium thinking, or an equivalent model that offers reliable structured output and tool use while remaining fast and cost-effective. This balance is especially important for consolidation, which may process a substantial conversation history and make several model calls per pass.
Supported channels: Slack, Discord, Telegram, Signal, SMS, WhatsApp Cloud, WhatsApp Personal.
Ontology-backed memory
- Add a complete PKM memory backend with user-scoped entities, aliases, attributes, relationships, temporal facts, source references, short-term memories, and reusable playbooks.
- Introduce staged background consolidation that extracts grounded facts from conversations, resolves duplicate identities, reconciles attributes and relationships, classifies entities, proposes ontology terms, and publishes deterministic entity pages.
- Build effective user ontologies from bundled releases and user-owned overlays, including lifecycle management, validation, reasoning, catalogue search, inspection, and SPARQL queries.
- Ground agent-derived memories in supporting tool execution evidence and retain transcripts, checkpoints, usage statistics, and recovery state throughout consolidation.
- Store PKM content in a user-scoped Markdown vault with page projection, rename, synchronization, reset, and recovery support.
- Add agent tools for remembering and searching memories, reading entities and playbooks, inspecting ontology terms, querying SPARQL, and retrieving cited source or tool evidence.
- Add scheduled consolidation sweeps with bounded concurrency, retries, durable checkpoints, transactional writes, cleanup, and recovery after interrupted work.
- Preserve the existing basic memory backend for upgrades, with configurable backend selection and a dedicated model group for background memory work.
Memory experience
- Add a Memory page with an interactive entity graph, inspector, page search, navigation, synchronization controls, and an empty-state experience.
- Expose consolidation status and history so users can see active, completed, and failed memory processing runs.
- Add user and administrator memory settings for selecting the backend and configuring PKM behavior.
- Render memory-search activity in chat and expose supporting citations and recall details.
- Improve graph visibility, page loading, origin filtering, usage-based browsing, and sandbox path handling.
- Hide neutral memory dispositions and canonicalize existing entity types.
- Retry conflicting entity writes and serialize initial ontology writes to improve consistency.
Inference and conversation handling
- Introduce typed structured inference with schema validation, repair conversations, retry classification, provider fallbacks, and optional request tracing.
- Centralize prompt assembly and add a system agent capable of running background inference without an active chat.
- Route inference through typed runtime model-provider configuration and Rig-backed model discovery.
- Support the OpenAI Responses API alongside Chat Completions, with provider metadata supplying the default where available.
- Preserve provider reasoning data and agent text surrounding internal tool calls.
- Add rolling chat-history compaction with durable summaries to keep long conversations within model context limits.
- Batch streamed chat updates to reduce frontend rendering overhead.
- Increase the token allowance used when generating conversation titles.
Channels and lifecycle
- Replace the monolithic channel manager with independently supervised channel workers and a unified lifecycle signal stream.
- Automatically reconnect failed channels according to their retry policy.
- Add enable, disable, reconnect, and reconnecting-state controls to the Channels interface.
- Migrate existing channel records with an explicit enabled state and clear stale connection statuses.
- Run the Signal worker in the correct local task context for more reliable operation.
- Correctly manage channel-owned spaces during deletion and include archived chats when deleting spaces.
Models and settings
- Improve model-group creation and provider navigation in the web interface.
- Persist disabled providers and removed model groups correctly.
- Expose provider and memory configuration through the agent and settings APIs.
- Standardize provider-specific parameters across Anthropic, OpenAI, Gemini, Ollama, and OpenAI-compatible services.
- Improve inference retry behavior, malformed tool-call argument repair, and required-tool validation.
Build, containers, and development
- Add support for multiple container runtimes through a shared container command wrapper.
- Improve container development performance with persistent compiler caching and more efficient build layers.
- Move the development image to Python 3.12 and align it with the production interpreter.
- Upgrade the Rust toolchain to 1.97.1 and pin Node.js 24.19.0.
- Fetch bundled ontology releases during image builds while keeping user ontologies on persistent storage.
- Refresh Rust, frontend, Python, and system dependencies.
- Automate pinned binary-tool updates and enforce workspace-wide Rust formatting.
- Improve end-to-end test isolation and reliability across PKM, MCP, sandbox, database, and server lifecycle coverage.
Reliability and bug fixes
- Preserve tool-turn text and provider reasoning content when reconstructing messages.
- Fix app process-group termination before supervised restarts.
- Stabilize MCP end-to-end tests and database runtime initialization.
- Improve Landlock sandbox coverage, workspace permission isolation, and path canonicalization.
- Fix memory graph navigation, visibility, empty states, and external-origin presentation.
- Retain test line information in builds for more useful failure diagnostics.
- Add extensive transactional, recovery, synchronization, consolidation, ontology, channel, and structured-inference test coverage.
Get started
- Quickstart: bring up the stack with Docker Compose in minutes
- Screenshots: see the platform in action
- Documentation: full guides and reference