v0.12.0-rc1 Prerelease

AD4M v0.12.0-rc1 — Holochain 0.7 (Iroh Transport) + Multi-User Support

This is a release candidate for v0.12.0 — a major milestone with two transformative features landing together.

🚀 Holochain 0.7 with Iroh Transport

The networking layer has been completely overhauled. Holochain 0.7 replaces the previous transport with Iroh — a modern, efficient peer-to-peer networking stack built on QUIC.

What this means in practice:

• Faster peer discovery and more reliable connections
• Lower latency for neighbourhood sync operations
• Better NAT traversal — fewer connectivity issues behind firewalls
• Reduced bandwidth and more efficient data transfer

This is the same Holochain upgrade we've been field-testing since v0.11.2-dev, now battle-hardened and ready for wider use.

👥 Multi-User Support

AD4M can now serve multiple users from a single node. This is the foundation for hosted deployments — run one AD4M instance and let multiple people connect with their own identities, perspectives, and data.

Key capabilities:

• Per-user isolation — each user gets their own perspectives, languages, and Holochain agent keys
• Email-based authentication — managed users sign up with email verification, get their own DID
• Admin/user separation — the main agent administers the node, managed users have scoped access
• Capability-based permissions — fine-grained control over what each user can access
• Notification system — migrated to SurrealDB for multi-user scalability

This unlocks the path to AD4M-as-a-service: host a node, invite your community, and everyone participates in neighbourhoods with full agent sovereignty.

What Else Is New

Stability & Performance

• Memory optimizations for multi-user workloads (#644)
• SurrealDB retry logic for read-write conflicts (#640)
• Deadlock fixes in perspective operations (#673)
• P-diff chunking to avoid Holochain entry size limits (#655) + panic fix (#669)
• Fixed link deserialization in p-diff sync (#674)

Security & Correctness

• Reject links with invalid or empty source/target/predicate URIs (#676)
• Admin-only perspective overview — now gated on launcher credential, not ALL_CAPABILITY (#675)
• Route neighbourhood signals to main agent when co-located with managed users (#680)

Developer Experience

• Updated @coasys/ad4m-test runner for v0.10.1+ compatibility (#670)
• Bootstrap seed support in test runner — no more stale bundle downloads (#678)
• Removed stale process kills from test runner (#686)
• Collection instance filtering and SurrealDB graph traversal improvements (#650)
• AD4M Connect updates (#646)
• SMTP configuration toggle (#647)
• Language tests migrated from Tryorama to Sweettest (#681)
• CI improvements: Docker builds, Node 18.x updates (#656, #657, #659)

What's Changed

• Multi-User/Agent Support for single AD4M nodes (1/3) by @lucksus in #629
• Hotfix blocked messages with Holochain debug fork by @lucksus in #642
• Email Verification System for Multi-User Authentication (2/3) by @lucksus in #643
• Memory Optimizations by @lucksus in #644
• Update Holochain to 0.7.0-dev.10-coasys by @lucksus in #645
• Ad4m Connect updates by @jhweir in #646
• Add SMTP Configuration Toggle by @lucksus in #647
• Migrate Notification System from Prolog to SurrealDB by @lucksus in #648
• Notification fixes for multi-user mode by @lucksus in #649
• Collection Instance Filtering and SurrealDB Graph Traversal Improvements by @jhweir in #650
• Switch SurrealDB from In-Memory to File-Based RocksDB Storage by @lucksus in #652
• Chunk regular diffs to avoid HC entry size limit by @lucksus in #655
• ci: Update language test workflows to Node 18.x by @data-coasys in #656
• ci: Use Docker container for npm-publish workflow by @data-coasys in #657
• ci: Optimize staging builds with Docker (Linux) and caching (macOS) by @data-coasys in #659
• fix: panic in p-diff-sync after diff chunking feature by @lucksus in #669
• fix(test-runner): update ad4m-test for v0.10.1+ compatibility by @HexaField in #670
• Deadlock Fixes by @data-coasys in #673
• P-Diff Chunking Fixes + Link Deserialization Fix by @data-coasys in #674
• fix: gate perspectives admin-overview on launcher credential, not ALL_CAPABILITY by @lucksus in #675
• fix: reject links with invalid or empty source/target/predicate URIs by @lucksus in #676
• fix(test-runner): use bootstrap seed instead of downloading stale language bundles by @HexaField in #678
• fix: route neighbourhood signals to main agent when co-located with managed users by @lucksus in #680
• Tryorama -> sweettest for agent, dm, and file-store language by @lucksus in #681
• Duplicate agent fix by @jhweir in #641
• Re-attempt surreal link additions up to 5 times if they hit read-write conflicts by @jhweir in #640
• fix: multi-user tests by @lucksus in #668
• fix(test-runner): remove stale process kills and perspect3vism refs by @HexaField in #686
• ci: Update Linux CI image to Ubuntu 24.04 for webkit2gtk-4.1 support by @data-bot-coasys in #684
• SHACL batch link operations by @HexaField in #685

Full Changelog: v0.11.1...v0.12.0-rc1

---

⚠️ Release Candidate — macOS binary coming soon. Please test and report issues!

v0.11.2-dev HC 0.7.0-dev with Iroh transport

This is development release from this branch #645, updating Holochain to a dev revision (0.7.0-dev.6) so we can field test the new Iroh transport.

v0.11.1

Holochain 0.6.0 + SurrealDB = Now Truly Lightning-Fast & Rock-Solid

We shipped v0.11.0 last Friday with the long-awaited Holochain 0.6.0 and our brand-new SurrealDB query engine — and the community immediately jumped in to test. The feedback was clear: the performance leap was massive, but a few edge-cases surfaced under real-world use - especially with nodes going offline again, we had to change the GetStrategy in our Holochain DNA.

Today we’re pushing v0.11.1 — the polished, production-ready version of the same big upgrade. All the speed, none of the rough edges.

What v0.11.0 + v0.11.1 deliver together

Holochain 0.6.0 (final, stable)

• No more random sync stalls or missing links that plagued 0.5.6 in large neighbourhoods
• Better networking, lower memory, and bullet-proof p-diff-sync validation

SurrealDB Query Engine

• Ad4mModel queries, filters, and real-time subscriptions are now orders of magnitude faster
• Flux conversations with 10 000+ messages + AI summaries scroll instantly
• CPU thrashing during heavy workloads is gone

What v0.11.1 fixes (hot on the heels of 0.11.0)

• One SurrealDB database per perspective → no more cross-perspective query bleed and better isolation (#636)
• Local-only get strategy for all DNAs → eliminates network timeouts during heavy reads (#638)
• Fixed language resolution with transform properties in SurrealDB queries (#637)
• Removed artificial delays in Ad4mModel.save() and batch.commit() → mutations are now truly instant (#639)

Result: the blazing-fast experience we promised in 0.11.0 is now 100 % stable under real-world usage.

tl;dr

Download the fixed, production-ready build right now:
https://github.com/coasys/ad4m/releases/tag/v0.11.1
(Look for the ADAM.Launcher… files)

Flux (latest stable) works perfectly:
https://app.fluxsocial.io
(dev build: https://fluxsocial-dev.netlify.app)

What's Changed (0.11.0 → 0.11.1)

• Refactor: Use Separate SurrealDB Database per Perspective by @lucksus in #636
• Set get strategy to local for all HC DNAs to avoid timeouts by @lucksus in #638
• Fix: Support non-literal resolveLanguage with transform in SurrealDB queries by @lucksus in #637
• Avoid delays in Ad4mModel.save() and batch.commit() by @lucksus in #639

Full combined changelog (0.10.1 → 0.11.1):
v0.10.1...v0.11.1

This is the ADAM we’ve been working toward all year — fast, private, and truly ready for scale.

Update now, open your biggest Flux neighbourhood, and feel the difference.

Happy building!
#ADAM #v011 #Holochain #SurrealDB

What's Changed

• Refactor: Use Separate SurrealDB Database per Perspective by @lucksus in #636
• Set get strategy to local for all HC DNAs to avoid timeouts by @lucksus in #638
• Fix: Support non-literal resolveLanguage with transform in SurrealDB queries by @lucksus in #637
• Avoid delays in Ad4mModel.save() and batch.commit() by @lucksus in #639

Full Changelog: v0.11.0...v0.11.1

v0.11.0

Holochain 0.6.0 + SurrealDB = Lightning-Fast, Rock-Solid ADAM

We’re thrilled to ship ADAM v0.11.0 — the fastest, most stable version we’ve ever released. This upgrade is all about raw performance and reliability, powered by two massive leaps forward:

Holochain 0.6.0 – now fully stable and running beautifully
SurrealDB Query Engine – adds to our existing Prolog query layer and brings orders-of-magnitude faster Ad4mModel operations

If you’ve ever felt a slowdown when working with large perspectives, long Flux conversations, or complex Ad4mModel queries — those days are over.

tl;dr

Binaries for Linux and macOS are below (look for the “ADAM.Launcher…” files). CLI binaries are perfect for servers or headless use.
Pair with the latest Flux at https://app.fluxsocial.io (or dev build: https://fluxsocial-dev.netlify.app)

What’s New & Why It Matters

Holochain 0.6.0 – Finally Here and Running Smoothly

After careful testing with rc.0 and rc.2, we’ve upgraded to the official Holochain 0.6.0 release.

• Better networking, reduced memory usage, and improved validation.
• Fixes long-standing p-diff-sync edge cases (parent commit validation → no more orphan diffs).
  See PR #630, PR #633

SurrealDB Query Engine – The Performance Game-Changer

This is the biggest internal upgrade since local AI in v0.10.

• Ad4mModel queries, filters, and subscriptions are now blazing fast — even with tens of thousands of entries.
• Real-time updates feel instant, no matter how big your perspective grows.
• Complex where clauses, pagination, and counts are handled natively by SurrealDB instead of in-memory Prolog scans.
• CPU thrashing during heavy async workloads is gone (PR #626).

Result: Flux conversations with thousands of messages + AI summaries stay buttery smooth.

Other Important Wins

• Ad4mModel.fromJSONSchema() – generate full models dynamically from a JSON schema (PR #628)
• Subscription stability fixes – no more missed updates (PR #634)
• Deno lockfile hygiene (PR #635)

What's Changed

• Async Runtime Performance Optimization: Fix CPU Thrashing Issues by @lucksus in #626
• p-diff-sync: Add validation to ensure parent commits exist before accepting PerspectiveDiffEntryReference by @lucksus in #627
• Add fromJSONSchema() method to Ad4mModel for dynamic model generation by @lucksus in #628
• Update to Holochain 0.6.0-rc.0 by @lucksus in #630
• Deno lockfile updates by @jhweir in #635
• Subscription fixes by @jhweir in #634
• Holochain bump to 0.6.0 by @lucksus in #633
• SurrealDB Query Engine Integration for blazing fast Ad4mModel operations by @lucksus in #632

Full Changelog: v0.10.1...v0.11.0

---

This is the ADAM we’ve been working toward: fast, stable, and ready for extended testing.
Update today and feel the difference — especially in Flux!

Let’s keep pushing the decentralized semantic web forward together.
Happy building!

v0.10.1

Unleashing AI-Powered, Distributed Semantics for All

After a year of intensive development and refinement through five release candidates, we're thrilled to announce ADAM v0.10.1—our first stable production release since v0.9.0 (March 23, 2024). This version represents a quantum leap toward the Synergy Engine: a decentralized, semantic network where ideas connect intelligently across communities. Built on feedback from our dedicated testers, v0.10.1 integrates local AI for privacy-first processing, upgrades to Holochain for robust distribution, and powerful tools like Ad4mModel for building dynamic apps. Flux, our flagship app, showcases this in action with real-time semantic conversations. Whether you're a developer, early adopter, or investor, this release empowers you to create and query meaningful connections in a truly distributed world. Dive in, test thoroughly, and let's shape the future together! 🚀

tl;dr:

Binaries for Linux and macOS are available below—look for the "ADAM.Launcher..." files. CLI binaries suit servers or developers. Pair with the latest Flux at: https://app.fluxsocial.io/.

What’s New?

Local AI Integration: Privacy-First Intelligence

ADAM now embeds local AI capabilities, letting apps harness LLMs, transcription, and embeddings without cloud dependency. Configure models once in ADAM’s settings—load from Hugging Face, local files, or APIs like Ollama—and all apps inherit them. Highlights include:

• Real-time audio transcription with Whisper models.
• Semantic analysis for grouping conversations into topics.
• Embedding vectors for intelligent search.
  See PRs #522, #504, #561, #562, #565, #566. Check the AI Integration Guide for setup.

Holochain Upgrades: Faster, More Reliable Distribution

Progressive updates from v0.3.4 to v0.5.6 bring massive performance boosts, better network handling, and our own relay server. Key optimizations:

• Diff-sync enhancements reduce DHT load and prevent failures (PRs #536, #608).
• Scoped Prolog pools for channel-specific queries, ignoring irrelevant data (PR #613).
• Fallback sync system with adaptive timing (30s on failure, 5min on success) to ensure link consistency (PR #625).
  Collaborated closely with the Holochain team—see PRs #542, #605, #618, #622.

Ad4mModel: Revolutionizing Data Handling

Evolved from SubjectEntity, Ad4mModel offers efficient ORM-like mapping with Prolog-powered queries:

• Complex where clauses compile to Prolog for real-time subscriptions (PRs #575, #573).
• Batch processing for atomic mutations, slashing latency (PR #587).
• Transform options, initial values, and better hooks (PRs #580, #585, #588).
  Ideal for apps like Flux. Explore the Ad4mModel Docs.

Flux: Semantic Conversations in Action

Flux demonstrates the Synergy Engine:

• Private, real-time call transcription posted as messages.
• AI-driven topic grouping and subtopics.
• Embedding-based search for similar discussions in Neighbourhoods.
  Soon scaling network-wide. Try it and see semantics at work!
  

Enhanced Debugging and Dev Tools

• Interactive Perspective Selector with SDNA debugging (PR #624).
• Holochain metrics, agent info, and restart button (PR #610).
• Configurable runtime logging via UI (PR #616).
• DB JSON export/import for easy backups (PR #569).

Revamped Documentation

Overhauled docs at docs.ad4m.dev cover everything from Installation to advanced Batch Operations (PR #582).

Overview of Release Candidates

We iterated through five RCs since January 2025, incorporating community feedback:

• RC1 (Feb 21): Introduced local AI MVP and Holochain v0.3.6.
• RC2 (Jan 20, internal): Refined AI CRUD, debouncing, and pending diffs.
• RC3 (Apr 15): Added query subscriptions, batch processing, and docs overhaul.
• RC4 (Aug 26): Performance boosts with scoped Prolog and Holochain v0.5.5.
• RC5 (Sep 10): Critical link sync fix and Holochain v0.5.6.

These RCs ensured stability—thanks to all testers!

Changes Since v0.9.0

Added

• Prolog predicates for Flux notifications: agent_did/1, remove_html_tags/2, string_includes/2, literal_from_url/3, json_property/3 (PR #483)
• Triggered notifications via OS (PR #483)
• App notifications with Prolog query triggers and web hooks (PRs #475, #482)
• ADAM executor hosting service alpha (PR #474)
• TLS for app interface (PR #489)
• Local AI for LLMs, embeddings, and transcription (PR #522)
• AI model CRUD and UI management (PR #504)
• Remote AI API support (PR #547)
• Wildcards in CLI query-links (PR #551)
• AI client in PerspectiveProxy (PR #558)
• Complex where queries with subscriptions (PRs #575, #573)
• Pagination for queries (PR #578)
• Property transforms (PR #580)
• Initial values in createSubject (PR #585)
• Batch mutations (PR #587)
• Dual transcription streams with cleanup (PR #590)
• Embedding vector caching for Prolog (PR #595)
• Holochain debugging features (PR #610)
• Scoped Prolog pools (PR #613)
• Runtime log configuration UI (PR #616)
• Perspective Selector with SDNA debugging (PR #624)

Changed

• Runtime service partial Rust migration (PR #466)
• SDNA/SubjectClass performance optimizations (PR #480)
• Launcher and Connect UI/branding updates (PRs #501, #502)
• Literal handling client-side (PR #498)
• Holochain to v0.3.2-rc1 (PR #506), then v0.3.6 (PR #542), v0.5.2 (PR #605), v0.5.5 (PR #618), v0.5.6 (PR #622)
• Prolog updates to O(1) for additions (PR #510)
• SubjectEntity to Ad4mModel with improved decorators (PR #579)
• Docs overhaul (PR #582)
• useSubjects to useModel with subscriptions (PR #588)
• Subscription delays extended (PR #604)
• Bootstrap server to relay.ad4m.dev (PR #606)
• P-Diff-Sync merges diffs into refs (PR #608)
• Ad4mModel.save() uses batches (PR #609)

Fixed

• useModel entry preservation (PR #617)
• Prolog respawn on errors (PR #483)
• Bootstrap seed generation (PR #496)
• QR scanning in Connect (PR #497)
• create_subject race conditions (PR #520)
• Subject Classes for non-progenitor agents (PR #521)
• Idle CPU usage (PR #531)
• Commit debouncing (PR #533)
• Long literals (PR #538)
• Missing links in hooks (PRs #540, #554)
• Failed commits in p-diff-sync (PR #551)
• Entry size limits (PR #553)
• AI task errors (PR #559)
• Link removals (PR #563)
• Local agent detection (PR #570)
• Signal handler losses (PR #576)
• Flaky subscriptions (PR #589)
• Missing languages on get (PR #594)
• Content escaping (PR #597)
• Passphrase handling (PR #596)
• Linux buttons (PR #599)
• Subscription setups (PR #601)
• Port crashes (PR #602)
• Ad4mModel subclassing (PR #603)
• Prolog race/deadlocks (PR #607)
• Expression parsing (PR #614)
• Missing neighbourhood links (PR #625)
• Link deduplication (PR #619)
• Special characters (PR #620)

All Merged PRs

For the full history:

• RC5: PRs #619–#625
• RC4: PRs #599–#618
• RC3: PRs #573–#597
• RC2 (internal): PRs #522–#554
• RC1: PRs #556–#570
• Plus foundational PRs #466–#521 from post-v0.9.0.

New Contributors

A warm welcome to @jhweir, @saul-jb, @niko-ng, and @HexaField! Your contributions made this release shine. 🌟

Full Changelog: v0.9.0...v0.10.1

@lucksus

v0.10.1 Release Candidate 5 - Final Touches

We're back with a revised Release Candidate 5 for v0.10.1, addressing a few lingering bugs from RC-4, upgrading Holochain to 0.5.6, and including a critical fix for missing links in new neighbourhoods. After thorough testing with Flux, we’ve made targeted improvements to handle edge cases, enhance debugging tools, and ensure robust link synchronization across agents. This release brings us closer to a stable v0.10.1. Please test RC-5 and share your feedback to help us finalize this big release v0.10.1! 🙌

tl;dr:

Binaries for Linux and macOS are available below—look for the "ADAM.Launcher..." files. CLI binaries are suited for servers or developers. Test this RC with the latest Flux dev build at: https://fluxsocial-dev.netlify.app/

What’s New?

Critical Fix: Link Synchronization in New Neighbourhoods

Resolved an issue where links created immediately after neighbourhood creation (e.g., by Flux) were visible only to the creating agent and missing for others who joined. We now ensure these early links are properly committed to the link language during creation and added a robust fallback sync system with adaptive timing (30s on failure, 5min on success) to catch any missed links. This ensures all agents see a consistent view of shared links. See PR #625.

Holochain Upgraded to 0.5.6

Bumped Holochain to 0.5.6 for improved stability and performance in distributed data handling, building on our collaboration with the Holochain team. See PR #622.

Prolog and Ad4mModel Bug Fixes

Fixed handling of special characters in Prolog queries and Ad4mModel to ensure smooth data processing, particularly in complex Flux scenarios with diverse inputs. Check out PR #620.

Enhanced SDNA and Link Handling

Added support for multiple SDNA classes with the same name by the same author, increasing developer flexibility. Implemented link deduplication to reduce redundant data in perspectives, optimizing sync efficiency. Details in PR #621 and PR #619.

Improved Debugging Tools

Introduced an interactive Perspective Selector with enhanced SDNA debugging, making it easier for developers to inspect and troubleshoot perspective data in apps like Flux. See PR #624.

What's Changed

• Special character fix in prolog & ad4m model by @jhweir in #620
• Allow for multiple SDNA classes with same name by same author by @lucksus in #621
• Bump Holochain to 0.5.6 by @lucksus in #622
• Deduplicate links by @lucksus in #619
• Interactive Perspective Selector with Enhanced SDNA Debugging by @lucksus in #624
• Fix Missing Links in New Neighbourhoods + Add Fallback Sync System by @lucksus in #625

Full Changelog: v0.10.1-release-candidate-4...v0.10.1-release-candidate-5

v0.10.1 Release Candidate 4 - Boosting Stability and Performance on the Road to the Synergy Engine

v0.10.1 Release Candidate 4 - Boosting Stability and Performance on the Road to the Synergy Engine

We’re excited to release the fourth (and final?) candidate for v0.10.1, incorporating critical fixes, performance optimizations, and upgrades based on ongoing testing and feedback from RC-3.

Basically: we used the last RC with Flux and had many several hour long conversations which produced a lot of transcribed messages and AI summaries and drove our systems to newly identified capacity limits. So, this build focuses on enhancing reliability in high-load scenarios. Main improvements being:

• Holochain: we jumped from very old 0.3.4 to 0.5.5, which yields an impressive performance boost!
• Prolog: we introduced automatic spawning, updating of and routing to scoped engine pools which limits the size of the fact databases when perspectives grow large. Enabling many long conversations in Flux without prohibitive performance degradation

It brings us a huge step closer to a robust, distributed semantic network. We encourage our community to test RC-4 and share your insights to help polish the final v0.10.1 release.

tl;dr:

Binaries for Linux and macOS are available below—look for the "ADAM.Launcher..." files. CLI binaries are suited for servers or developers. The latest development build of Flux for testing this RC is hosted at: https://fluxsocial-dev.netlify.app/

What’s New?

Holochain and Deno Upgrades

We've progressively upgraded Holochain through versions 0.5.2, 0.5.3-rc0, 0.5.4, and finally to 0.5.5, alongside Deno to 0.212.0. These updates improve compatibility, stability, and performance in distributed data handling, including better network configuration and dependency alignment. We cooperated closely with the Holochain team and found new, relevant issues in our internal testing rounds which have all been fixed by the Holochain team in the latest 0.5.5 release which is included in this build. Additionally, we've switched the Holochain bootstrap server to our own relay for enhanced reliability. See details in PR #605, PR #606, PR #612, PR #615, and PR #618.

Prolog Query Enhancements

Significant improvements to Prolog query subscriptions and inference speed. Main effort being scoped engine pools for faster channel-specific queries (ignoring irrelevant data from other channels to reduce the size of the fact database for queries that clearly are scoped to one channel, or root node in the Perspective graph) PR #613

Other Prolog related improvements includ hardened setup with delayed initializations, fixes for race conditions and deadlocks, and scoped engine pools . These changes ensure more reliable real-time updates and reduce overhead in large perspectives. Explore the updates in PR #601, PR #604, PR #607.

Ad4mModel Improvements

Refinements to the Ad4mModel for better usability and efficiency, such as enabling usage without subclassing, parsing resolved expression data, integrating batch mode in save operations to minimize commits, and preserving entry instances in hooks. These fixes enhance data consistency and performance for apps like Flux. Check out PR #603, PR #609, PR #614, and PR #617.

Sync and Debugging Optimizations

Optimized perspective diff synchronization by merging diffs into references, reducing DHT load and ensuring atomicity to prevent sync failures. New Holochain debugging features include network metrics display, agent info, and a restart button without full app reload. Configurable logging adds runtime log level adjustments via UI for better diagnostics. Details in PR #608, PR #610, and PR #616.

Stability Fixes and Minor Updates

Addressed various crashes (e.g., port collisions, Flux channel creation with plugins) and UI issues (e.g., Linux button functionality). Added tests for neighbourhood subscriptions and fixed minor typos. These contribute to overall robustness. Refer to PR #599, PR #600, PR #602, PR #577, and PR #586.

What's Changed

• Linux button fix by @lucksus in #599
• Fix crash when creating Flux channel with all Plugins by @lucksus in #600
• Harden Prolog query subscription setup by @lucksus in #601
• Fix port collision crash by @lucksus in #602
• Send more inits for query subscriptions (10s - 25s) by @lucksus in #604
• Feature/fix usage of ad4m model without subclass by @lucksus in #603
• Upgrade Deno and Holochain to 0.212.0 and 0.5.2 respectively by @lucksus in #605
• Set HC bootstrap server to our own relay by @lucksus in #606
• Fix race condition in prolog query subscriptions and resolve deadlock issues by @lucksus in #607
• Use batch mode inside Ad4mModel.save() by @lucksus in #609
• P-Diff-Sync: merge perspective diffs into refs -> only one HC entry per commit by @lucksus in #608
• Holochain debugging features by @lucksus in #610
• Update to HC 0.5.3-rc0 by @lucksus in #612
• Parse resolved expression data in Ad4mModel by @jhweir in #614
• Scoped Prolog engine-pools for fast inference inside a channel (ignoring other channels' data) by @lucksus in #613
• Upgrade Holochain to 0.5.4 by @lucksus in #615
• fix typo by @niko-ng in #586
• add test for multiple neighbourhood apollo subscriptions by @HexaField in #577
• Configurable logging by @lucksus in #616
• Entry instances preserved in useModel hook when making the base expre… by @jhweir in #617
• Upgrade Holochain to 0.5.5 by @lucksus in #618

New Contributors

• @niko-ng made their first contribution in #586

Full Changelog: v0.10.1-release-candidate-3...v0.10.1-release-candidate-4

v0.10.1 Release Candidate 3 - Refining the Path to the Synergy Engine

v0.10.1 Release Candidate 3 - Refining the Path to the Synergy Engine

We’re pleased to announce the third release candidate for v0.10.1, building on a year of progress since our last public release (v0.9.0, March 23, 2024). This RC incorporates feedback from RC-1’s public testing, addressing performance bottlenecks and introducing powerful new features. It’s a significant step toward our vision of a distributed, semantic network powered by the Synergy Engine. We invite our community—friends, early adopters, and investors—to test RC-3 and help us finalize v0.10.1.

tl;dr:

Binaries for Linux and macOS below. You want the "ADAM.Launcher..." files. CLI binaries are for servers or devs. Latest development build of Flux which you'll need to test this RC is hosted at: https://fluxsocial-dev.netlify.app/

Walkthrough video: https://www.youtube.com/watch?v=Zmhhx3kRG0U

What’s New?

Local AI Integration in ADAM

ADAM now supports integrated local AI, enabling apps to leverage large language models (LLMs), transcription, and embedding tasks through straightforward interfaces. Configure your preferred local or remote models once in ADAM’s settings, and all apps inherit that setup—privacy-first and highly flexible. Load models from Hugging Face or local files with ease. Explore the details in PR #522 and PR #504, or dive into the AI Integration Guide.

Holochain Upgraded

We’ve adopted Holochain v0.3.6 (PR #542), enhancing performance and reliability for distributed data. Combined with optimizations like improved diff-sync (PR #536) and faster Prolog updates (PR #510), ADAM’s foundation is stronger than ever.

Flux: Semantic Conversations

Our flagship app, Flux, demonstrates ADAM’s potential:

• Real-time, private transcription of calls using local AI.
• Transcriptions posted as chat messages, forming a conversation log.
• Local LLMs analyze discussions, grouping them into topics and subtopics.
• Embedding vectors enable searching for similar conversations within a Neighbourhood.
  

Flux is a working prototype of the Synergy Engine’s semantic query capabilities, currently local to Neighbourhoods and soon scalable network-wide. Try it out and see the future taking shape.

Ad4mModel: Faster and Smarter

than SubjectEntity
Feedback from RC-1 testing drove significant improvements to revamp our Object-Graph-Mapping approach and rename it into Ad4mModel:

• First, rewrote the existing query mechanism to allow for complex query combinations, leveraging the underlying Prolog engine, compiling these queries down to Prolog (PR #575)
• Then, Prolog Query Subscriptions (PR #573) were added to allow push updates from AD4M for all of these complex queries - getting real-time updates when perspective query results change, enabling dynamic apps to react instantly (e.g., Flux updating topic groups as new messages arrive). See the Ad4mModel Docs for details.
• Batch Processing (PR #587): Execute multiple operations—like creating or updating data—in a single transaction, slashing latency for apps like Flux handling high-volume transcriptions. Learn more in the Batch Operations Guide.

Revamped Documentation

Our documentation has been overhauled (PR #582) to support users and developers. Key resources:
Getting Started for first-time users.

AI Integration for configuring models.

Ad4mModel for building apps.

Batch Operations for optimizing data workflows.

Other Improvements

• Dynamic AI Models: Add DeepSeek, Qwen, or Whisper models at runtime (PR #561, PR #562).
• Developer Tools: Enhanced SubjectEntity queries with where clauses (PR #560) and voice activation for transcription streams (PR #566).
  -Ad4m Connect: Mobile QR code scanning and UI polish (PR #497, PR #509).

Why This Matters

This release advances our Synergy Engine vision: a network where semantic queries connect ideas across distributed communities. Flux’s real-time analysis and Ad4mModel’s performance gains are tangible milestones, refined by RC-1 feedback. RC-3 brings us closer to a scalable, decentralized future, and your testing will ensure v0.10.1 delivers.

All merged PRs

What's Changed (since RC-1)

• Prolog query subscriptions by @lucksus in #573
• Fix multiple synchronous addSignalHandler calls losing the first handler by @HexaField in #576
• Enhanced SubjectEntity Query System by @jhweir in #575
• Pagination and count for new SubjectEntity queries by @lucksus in #578
• Rename SubjectEntity to Ad4mModel and adjust decorator names by @lucksus in #579
• Add transform option to Ad4mModel properties by @lucksus in #580
• Fix snake_case <> CamelCase matching for perspectiveKeepAliveQuery by @lucksus in #581
• Ad4m model class name override & use of baseExpression in where queries by @jhweir in #584
• Documentation upgrade by @lucksus in #582
• createSubject() with initial values, used in Ad4mModel.save() to avoid link removals and Prolog fact reloads by @lucksus in #585
• Atomic batch processing of perspective mutations by @lucksus in #587
• New useModel hooks set up by @jhweir in #588
• Fix flakiness of prolog subscription queries by @lucksus in #589
• Multiple Transcription Streams Support & Automatic Stream Cleanup by @lucksus in #590
• Improve Query Subscription Management and Resource Cleanup by @lucksus in #591
• Improved perspective handling in Vue useModel hook by @jhweir in #592
• Install missing languge during expression get by @lucksus in #594
• Multiple Prolog engines per Perspective for parallel queries by @lucksus in #593
• Exclude huge embedding vector strings from Prolog, replace with small cache ID by @lucksus in #595
• Passphrase repromt on setup and error on failed unlock by @lucksus in #596
• Fix escaping of expression content by @lucksus in #597

Full Changelog since RC-1: v0.10.1-release-candidate-1...v0.10.1-release-candidate-3

Changes between last public release-candidate-1 and internal rc2 (v0.10.1-rc2 Jan 20th 2025)

• Agent list handling imrovements and other cleanup by @lucksus in #556
• SubjectEntity improvements by @lucksus in #557
• Add ai client to PerspectiveProxy by @lucksus in #558
• Fix error after spawning AI task by @lucksus in #559
• Add where to SubjectEntity's query() by @lucksus in #560
• Add DeepSeek and Qwen models, update Kalosm/candle by @lucksus in #561
• Fix link-removal problems with proper implementation by @lucksus in #563
• Add models at run-time directly from Hugginface repository by @lucksus in #562
• Whisper model size selection by @lucksus in #564
• Add AI models from local files by @lucksus in #565
• Optional voice activation params for ad4m.ai.open_transcription_stream() by @lucksus in #566
• Neighbourhood signal broadcast loopback by @lucksus in https://github.com/c...

v0.10.1 Prerelease (final RC for wider testing)

v0.10.1 (RC) - A Leap Toward the Synergy Engine

Hey friends, supporters, and early adopters! It’s been nearly a year since our last public release (v0.9.0, March 23, 2024), and we’re thrilled to share this release candidate for v0.10.1. This is a big step forward for ADAM and our vision of a distributed, semantic future with the Synergy Engine. We’ve packed a ton of innovation into this release—especially with local AI integration and a rock-solid Holochain update—and we’d love your help testing it over the next few days before we finalize v0.10.1.

tl;dr:

Binaries for Linux and macOS below. You want the "ADAM.Launcher..." files. CLI binaries are for servers or devs. Latest development build of Flux which you'll need to test this RC is hosted at: https://fluxsocial-dev.netlify.app/

What’s New?

A f***ing lot! Here are the highlights:

Local AI Comes to ADAM

The star of this release is integrated local AI in ADAM. Now, ADAM apps can seamlessly tap into large language models (LLMs), transcription, and embedding tasks via simple interface methods. Set your preferred local or remote models once in ADAM’s settings, and every app you run inherits that AI config. Privacy-first, powerful, and flexible—whether you’re loading models from Hugging Face or your own files. Curious? Check out PR #522 and PR #504 for the details.

Holochain Gets an Upgrade

We’ve updated to Holochain v0.3.6 (PR #542), and it’s running smoother than ever. This brings better performance, reliability, and a foundation for our distributed network goals. Paired with fixes like improved diff-sync (PR #536) and faster Prolog updates (PR #510), ADAM is more robust than ever.

Flux: Your Conversations, Supercharged

Our flagship ADAM app, Flux, showcases what’s possible with this release. Imagine joining a call where:

• Local AI transcribes everything privately in real-time.
• Transcriptions post as chat messages, creating a live conversation log.
• Local LLMs analyze the discussion, grouping it into topics and subtopics.
• Embedding vectors let you search for similar conversations within your Neighbourhood.
  

This is a working MVP of the Synergy Engine use-case—semantic queries across a local group today, and soon, the entire ADAM network tomorrow. We can’t wait for you to try it!

Other Highlights

• Dynamic AI Models: Add DeepSeek, Qwen, or Whisper models at runtime (PR #561, PR #562).
• Better Developer Tools: New where clauses for SubjectEntity queries (PR #560) and voice activation options for transcription streams (PR #566).
• Ad4m Connect Polish: QR code scanning on mobile and UI tweaks (PR #497, PR #509).

Why This Matters

This release isn’t just a feature drop—it’s a milestone. The local AI in ADAM and the semantic magic in Flux are proof-of-concept for the Synergy Engine: a future where you can query and connect ideas across the entire distributed ADAM network, not just your local Neighbourhood. We’re getting close, and your feedback now will shape the final v0.10.1.

Join the Testing Party!

We’re releasing this candidate to give our community—YOU—a chance to kick the tires before the official v0.10.1 drop. A quick heads-up on builds:

• Windows: No binary yet (we’re working on it!).
• Linux: Requires a recent distro with new glibc (see Tauri issue #9039). We’ve got CUDA and non-CUDA builds—use CUDA for local LLMs with a beefy GPU (min. 6GB VRAM, 8GB+ recommended).
• macOS: Metal-powered and ready for all M1+ Macs. Intel build in progress.

Download it ("ADAM Launcher" files), play with Flux (use the development deploy: https://fluxsocial-dev.netlify.app/), tweak some AI settings, and let us know what works (or doesn’t) via GitHub Issues. Focus on Flux’s transcription, topic grouping, and search—those are the stars we want to shine.

Shoutouts

Big thanks to our contributors, including new faces
@jhweir and @saul-jb, and core team members
@lucksus and @fayeed. This is a collective win!
Let’s make this final push to v0.10.1 awesome together. Happy testing!

All merged PRs

Changes since last public RC (v0.10.1-rc2 Jan 20th 2025)

• Agent list handling imrovements and other cleanup by @lucksus in #556
• SubjectEntity improvements by @lucksus in #557
• Add ai client to PerspectiveProxy by @lucksus in #558
• Fix error after spawning AI task by @lucksus in #559
• Add where to SubjectEntity's query() by @lucksus in #560
• Add DeepSeek and Qwen models, update Kalosm/candle by @lucksus in #561
• Fix link-removal problems with proper implementation by @lucksus in #563
• Add models at run-time directly from Hugginface repository by @lucksus in #562
• Whisper model size selection by @lucksus in #564
• Add AI models from local files by @lucksus in #565
• Optional voice activation params for ad4m.ai.open_transcription_stream() by @lucksus in #566
• Neighbourhood signal broadcast loopback by @lucksus in #568
• Ad4m Connect updates by @jhweir in #570
• DB JSON export/import by @lucksus in #569

Full diff: v0.10.1-rc2...v0.10.1-release-candidate-1

Changes between that and retracted release v0.10.0 (Dec 23rd 2024 -- Jan 20th 2025)

• Detect and rerun failed commits in p-diff-sync by @lucksus in #551
• New useSubjects hook filter to reduce unnecissary re-renders by @jhweir in #552
• Prevent p-diff-sync diff entries from exceeding HC entry size limit of 4MB by @lucksus in #553
• Fix race-condition in useSubjects by @lucksus in #554

Full diff: v0.10.0...v0.10.1-rc2

Changes between last full public release 0.9.0 and retracted 0.10.0 (Mar 23rd 2024 -- Dec 23rd 2024)

• Runtime service rust implementation by @fayeed in #466
• Agent & entanglement rust implementation by @fayeed in #469
• Complete build description in readme by @lucksus in #473
• Host root by @fayeed in #474
• App Notifications through SDNA triggers 1 - base implementation by @lucksus in #475
• Improved SDNA/Subject Class performance by @lucksus in #480
• Agent json parsing fixed by @fayeed in #481
• Auto reconnect websocket in Ad4mConnect by @lucksus in #484
• Connect hosting fixes by @fayeed in #485
• chore: Remove unnecessary console.log statement in Ad4mConnect class by @fayeed in #486
• Notifications 2 - web hook push by @lucksus in #482
• Update HolochainService debug log to truncate payload string by @fayeed in #488
• Update Ad4mConnect setUrl method to use provided URL instead of hardcoded value by @fayeed in #490
• TLS encryption for GraphQL app interface websocket by @lucksus in #489
• Notification Prolog helper predicates and OS notification push by @fayeed in #483
• Added log-holochain-metrics argument to CLI by @fayeed in #494
• Notifications Fix by @lucksus in #495
• Fix bootstrap publishing by @fayeed in #496
• Connect capacitor update to include qr code scanning on mobile by @fayeed in #497
• Process expression.get() on client for literals (no signature checking) by @lucksus ...

ADAM (Launcher & CLI binaries) v0.10.1-rc2 Prerelease

Pre-release for extended testing

Won't work with Flux 0.9, only with unreleased development branches of Flux (like coasys/flux#486, deployed here: https://deploy-preview-486--fluxsocial-dev.netlify.app/#/)

These are all the changes since last public RC v0.10.0-rc9. Version 0.10.0 got retracted after NPM packages got already released so next stable release will be v0.10.1 which will show all changes between 0.9.0 and 0.10.1 (changelog of 0.10.0-rc9, this version and future chagnes)

What's Changed

• Local AI model inference by @lucksus in #522
• Don't build any Rust code on staging npm publish by @lucksus in #527
• Fix vue hook private member error. by @saul-jb in #529
• Change CLI default port to 12000 by @lucksus in #530
• Add pre-push and setup-hooks scripts for code formatting and linting for windows by @fayeed in #523
• Disable V8's optimisation and turbo to avoid high CPU usage when idling by @lucksus in #531
• Ad4mConnect: change authState to 'authenticated' on successful connect by @lucksus in #532
• AI Model interface CRUD by @lucksus in #504
• Debounce link language commits - safe guard against commit-strom by @lucksus in #533
• Refactor pending diff handling: don't miss any diffs by @lucksus in #536
• Fix parsing of long literal properties by @lucksus in #538
• Timeout fallback added to linkAdded function in useSubjects hook by @jhweir in #540
• Tune model config and selection & other misc. clean-ups and fixes by @lucksus in #539
• AI setup wizard providing more info about models being installed by @jhweir in #541
• CI fixes and improvements by @lucksus in #544
• Fix ad4m-connect bug introduced in PR#532 by @lucksus in #543
• Upgrade to Holochain v0.3.6 by @lucksus in #542
• Notification-request-dialog improvements by @lucksus in #545
• Bug fix: allEntries readded to linkAdded function in useSubjects hook by @jhweir in #546
• New bootstrap Holochain agent-info by @lucksus in #548
• Remote AI API improvements for Ollama by @lucksus in #547
• Detect and rerun failed commits in p-diff-sync by @lucksus in #551
• New useSubjects hook filter to reduce unnecissary re-renders by @jhweir in #552
• Prevent p-diff-sync diff entries from exceeding HC entry size limit of 4MB by @lucksus in #553
• Fix race-condition in useSubjects by @lucksus in #554

New Contributors

• @saul-jb made their first contribution in #529

Full Changelog: v0.10.0-rc9...v0.10.1-rc2