Skip to content

1.21.0 — Finance Alpha research terminal: next-open paper execution, auditable accounting, risk controls, CSV import, browser reports and exact replay; corrected legacy cash and P&L.

1.20.1 — Planning lab reliability: JSON downloads, clear failure recovery, strict input bounds, source provenance and corrected macOS setup.

1.20.0 — Evidence & Planning Lab: runnable MuZero/MCTS/LLM integration, exact citation validation, safe local launch and measured simulator comparisons.

1.19.0 — MuZero Planning Lab: actual recurrent learning, corrected reward-aware search, measured comparisons, safe checkpoints and verified local/browser/container paths.

1.18.0 — Curriculum Lab: independently solved tasks, adaptive curricula, complete lineage, review gates, reproducible evidence and repaired legacy execution/accounting boundaries.

1.22.0 — Sovereign Workbench

  • Added a native, review-gated portfolio/schedule/evidence workflow with signed exports and recovery.
  • Added a local authenticated console, recorded public evidence and explicit operating boundaries.
  • Preserved the complete original deployment, standalone HTML, gallery presentation and diagrams.
  • See release notes.

1.13.1 — 2026-09-28

1.14.0 — Governance Workbench

  • Add nine inspectable proposal gates, five adversarial cases and exact Python/browser replay.
  • Export five-file evidence bundles and input-bound Ascension verification jobs.
  • Correct the legacy update-rate explanation and repair optional-runtime fallback.
  • Preserve original manuscripts, diagrams, presentations and archived research content.

Business 3 now presents a compact research collection, a styled original replay with explicit units, and a dossier-backed candidate value landscape. Original diagrams, media and controls are retained. Browser release gates verify the new visuals and original data.

1.13.0 — 2026-09-27

  • Deliver Business 3 Enterprise Studio: exact constrained portfolio planning, five editable cases, downside stress and seven portable evidence files with Python/browser parity.
  • Export reviewable goal/metric/bounty jobs accepted by Ascension; preserve the existing full lifecycle and clearly distinguish unsubmitted plans from deployed enterprises.
  • Repair the Python and console entry points, offline container launcher and executable Colab notebook; add Business 3 to all 15 finite installed-wheel launch contracts.
  • Make optional research inference and integrations explicit, restore caller settings, close clients correctly and reject unsupported proof or model-update claims.
  • Preserve original flowcharts, PDF/PPTX and historical research material; require exact public assets, notebook execution, real Docker isolation, accessibility and offline acceptance before publication.

1.12.2 — 2026-09-27

  • Package all six demo CSV samples and 11 unchanged Insight scenario fixtures; run all 14 finite demos from the wheel outside the repository with Python network calls blocked.
  • Add read-only prerequisite checks, fail before launching incomplete installs, prevent output-directory module shadowing, and report output errors without losing data.
  • Pin finite offline examples to packaged inputs and provider-free settings; retain all advanced launch paths and parent credentials.
  • Add a complete demo walkthrough, refresh all 26 additive launch sections, and extend the installed-wheel acceptance matrix across Python 3.11–3.13.
  • Preserve every original demo, diagram, media asset and manuscript; retain the existing native/operator and local-EVM release gates.

1.12.1 — 2026-09-27

  • Reject contradictory signed approvals and boolean/float aliases in native result verification.
  • Repair the mirrored Protocol Desk and require exact published assets, receipt downloads, keyboard/accessibility and offline acceptance.
  • Include starting, factory and protocol guides in release assets and a linked operator-guide archive; reject dirty tracked release sources.
  • Bind public protocol acceptance to the exact release commit/version and retain all original paths, diagrams, media and manuscript bytes.

1.12.0 — 2026-09-27

  • Updated the Alpha-Factory entry point, operating guide, deployment paths and scope labels while preserving original flowcharts and media.
  • Added packaged examples and a native Ascension handoff: exact FusionPlan commitments, tamper checks, reviewed-result binding and externally anchored signature verification.
  • Unified cross-platform launchers, corrected source paths and Python support, added an operator-only preflight, and made failed/partial installations stop visibly.
  • Added native/JavaScript/Solidity interoperability and package preservation acceptance gates. No journal or contract migration.

1.11.0 — 2026-09-27

Add an integrated Ascension Solidity reference suite: ERC-721 Nova-Seeds, expiring validator risk oracle, MARK curve AMM, plan-bound Sovereign treasury, ENS/stake-gated auctions, reserved reviewer capacity, artifact-bound validation, exact $AGIALPHA payout burns, failure recovery and successor lineage. Add a native-agent/local-EVM acceptance run, public transaction desk and continuous validation. All existing contracts, papers, demos and capabilities remain. See the protocol guide.

Version 1.10.1 — Operator reliability audit

  • Remove the legacy wheel signature bypass and unify signing/verification formats.

  • Bound the full inference request, including continuously trickling responses.

  • Fence late worker failures from recovered replacement work.
  • Return normal authentication denial for malformed authorization values.
  • Enforce recoverable backup size, partial-write cleanup and completed-archive flushes.
  • Keep Decision Studio calculations and existing dossiers compatible.
  • See release notes and the operator guide.

Version 1.10.0 — Operational planning audit

  • Replace inventory-based service capacity with per-shift staffing, backlog, caps and cost accounting.
  • Prioritize job commitments, expose deadline slack, and keep infeasible schedules on HOLD.
  • Harden evidence numbers, prerequisite traversal, input bounds and versioned dossier replay.
  • Preserve keyboard focus and bind public Studio acceptance to the exact release commit.
  • See release notes and the method guide.

Version 1.9.0 — Practical decision workspaces

  • Replace public toy examples with eleven editable, useful operational cases and seven calculation engines.
  • Connect all 26 original entries to practical workspaces with distinct previews; preserve original implementations.
  • Add CSV inputs, decision briefs, plans, proof-job backlogs, verified dossiers and a matching Node CLI.
  • Expand native mission defaults and require browser export/replay, accessibility, mobile and offline acceptance.
  • Document exact algorithms, bounds, constructed data and evidence limits in the studio guide.

1.8.1 — Accessible experiment resets

Compounding Lab now clears stale prediction descriptions when scenarios or costs change. Browser acceptance checks both resets, alongside the existing manuscript, replay and release gates. See release notes.

1.8.0 — Final manuscript and Compounding Lab

The original 198-page manuscript, source provenance and all figures accompany the new future-task transfer experiment, complete Evidence Dockets and hardened native API. See release notes and the implementation map.

1.7.0 — Nova-Seeds Proof Bloom

Five connected guided experiences turn claims into bounded native Missions, replayed ProofBundles, explicit reviews and reversible Chronicle capability. The foundry adds exact benchmark reuse, transitive revocation, signed native returns, portable recovery and a new flagship Pages experience. See release notes and the field guide.

1.4.0 — usable browser workspace and complete demo catalog

A new GitHub Pages workspace connects checked browser missions, local ONNX generation, reviewed exports, optional memory, verification of signed agent results and a guided handoff to the installed agent. All original demos, documents and flywheels are retained. The complete launch catalog distinguishes working backends, simulations and preserved deployment templates. Dedicated browser and CPU checks cover the repaired demos. The exact tested site is published and included in the release archive. See release notes and demo evidence.

After 1.3.0 — live badge maintenance

Scope CI badges to main, use live watchdog and complete matrix results for Health/Smoke/Integration, and prevent unrelated Health runs from cancelling one another. Demo build badges now report real CI; research scope and linked validation replace static production and 100% coverage claims. All original paths, README prose and flywheels are preserved. Published 1.3.0 assets remain unchanged. See badge meanings.

1.3.0 — browser inference and release verification

Real pinned offline ONNX generation, memory-only browser credentials, CSP-compatible WebGL rendering, working direct AIGA launch, automatic nine-platform/Python smoke acceptance and main-scoped badges. See release notes. Original files and flywheels remain present.

1.2.0 — connected $AGIALPHA Agent

See release notes, capabilities and operator guide. Original content and flywheels are preserved.

See docs/DISCLAIMER_SNIPPET.md

Changelog

All notable changes to this project are documented in this file.

Breaking Changes Policy

Incompatible updates are announced in advance whenever possible and remain deprecated for at least one minor release. Each release includes a dedicated ### Breaking Changes section describing removed features or behavioural differences. Consult that section when upgrading.

Breaking Changes

Any incompatible API updates will be clearly listed under this heading for each release. Downstream users should consult this section when upgrading.

  • 1.1.0 – GraphMemory._fallback_query now returns stub data when both Neo4j and NetworkX are missing.
  • 1.0.3
  • check_env.py exits with an error if numpy or pandas are missing unless --allow-basic-fallback is used.
  • Removed alpha_asi_world_model_demo_v1.py.
  • Removed insight_browser_v1/dist/ from version control; run npm run build to regenerate the assets.

[Unreleased]

Breaking Changes

  • Adopted 18-decimal AGIALPHA token (0xa61a3b3a130a9c20768eebf97e21515a6046a1fa). Removed multi-token support and setToken functions. Existing deployments should scale token amounts by 1e12 when migrating from the previous 6-decimal token.
  • Core modules previously under src/ now live in alpha_factory_v1.core.
  • Update imports accordingly.
  • Monitoring, governance and archive helpers moved under alpha_factory_v1.core.
  • alpha-factory and edge_runner.py now print a short warning before startup.
  • Synced openai, openai-agents and uvicorn pins across requirements files and clarified why requests and rich differ between layers.
  • build_gallery_site.sh and deploy_gallery_pages.sh now run scripts/mirror_demo_pages.py so the GitHub Pages subdirectory mirror stays current.
  • Documented API_RATE_LIMIT, AGI_ISLAND_BACKENDS and ALERT_WEBHOOK_URL environment variables.
  • Added alpha_factory_v1/core/tools/analyse_backtrack.py for visualising archive backtracks.
  • Documented how to build a wheelhouse for offline installs and updated tests/README.md with the instructions.
  • Added scripts/build_offline_wheels.sh to gather wheels for all lock files.
  • Added requests to requirements-dev.txt for parity with other examples.
  • Updated @vitejs/plugin-vue to version 6 so the Docs workflow builds with Vite 7.
  • Removed outdated OPENAI_CONTEXT_WINDOW reference from the self-healing repo demo.
  • Documented how to add new policies in POLICY_RUNBOOK.md.
  • Each demo README now clarifies that __version__ denotes the demo revision, not the overall project release.
  • Clarified that transfer_test.evaluate_agent is a placeholder returning the archived score. Added a simple heuristic in MetaRefinementAgent that proposes a longer cycle period when logs show slow agent cycles.
  • Service worker cache version is now generated dynamically to ensure updates are picked up by browsers.
  • scripts/fetch_assets.py now downloads the Pyodide runtime from the official CDN and retrieves GPT-2 weights directly from Hugging Face.
  • Added FETCH_ASSETS_ATTEMPTS environment variable to control download retries.
  • Added FETCH_ASSETS_BACKOFF environment variable to adjust retry delay.
  • The browser bundle now defaults to the official Web3 Storage CDN. IPFS fallback has been removed.
  • Fixed internal documentation links so MkDocs builds without warnings.
  • The 📦 Browser Size workflow now caches pip and npm dependencies using the built-in features of actions/setup-python and actions/setup-node so repeated runs skip redundant downloads.
  • The workflow preinstalls numpy, pandas, pytest and PyYAML so the environment check passes reliably before running pre-commit.
  • Fixed the Build & Test GitHub Actions workflow so Docker tags always use lowercase repository names. The job now runs only when triggered manually by the repository owner.
  • Upgraded the Insight demo to streamlit 1.46 so NumPy 2.x installs cleanly.
  • The Docker image now installs rustc and cargo to build crates like blake3 on Python 3.13.
  • The workflow pushes version-specific Docker tags py311, py312, py313 and py314 for each build and signs all images with cosign.
  • ci.yml now caches Node dependencies using cache-dependency-path so the "Dependencies lock file is not found" warning no longer appears.

[0.1.0-alpha] - 2025-05-01

  • Initial alpha release.
  • Git tag v0.1.0-alpha.
  • This tag points at commit 0ff79a4f.
  • If cloning from a snapshot without tags, recreate it using: bash # Verify the release tag git tag -l v0.1.0-alpha # Recreate the tag when missing ./scripts/create_release_tag.sh HEAD git push origin v0.1.0-alpha
  • The package exposes alpha_factory_v1.__version__ = "0.1.0-alpha" at this release.

Breaking Changes

  • None.

[1.1.1] - 2025-08-01

  • CI workflow now runs via workflow_dispatch. The pipeline executes lint, mypy, the full Pytest matrix, Windows/macOS smoke tests, an offline docs build, the Docker build and pushes images on tags.
  • Lock files must be regenerated with pip-compile --allow-unsafe.
  • Requires Python 3.11–3.13 (<3.14) and Node.js 22.17.1.

[1.0.3] - 2025-07-10

Breaking Changes

  • check_env.py exits with an error if numpy or pandas are missing unless --allow-basic-fallback is used.
  • Removed alpha_asi_world_model_demo_v1.py.
  • Removed insight_browser_v1/dist/ from version control; run npm run build to regenerate the assets.
  • Added CI workflow running lint, type checks, tests and Docker build with automated image deployment on tags and rollback on failure. Metrics are exported via OpenTelemetry and can be viewed in Grafana or the Streamlit dashboard.
  • Clarified that numpy and pandas are required for realistic results and check_env.py now errors if they are missing unless --allow-basic-fallback is provided.
  • Documented ALPHA_ASI_* demo variables in README and AGENTS.md. and README (introduced in commit d0e1e2ed).

[1.1.0] - 2025-07-15

Breaking Changes

  • GraphMemory._fallback_query now returns stub data when both Neo4j and NetworkX are missing.

Added

  • CLI commands agents-status and replay for monitoring agents and replaying ledger events.
  • ADK and MCP adapters bridging the Alpha Data Kernel and Multi‑Chain Proxy.
  • Detailed design overview for the α‑AGI Insight demo under alpha_factory_v1/demos/alpha_agi_insight_v1/docs.
  • Expanded API documentation with curl and WebSocket examples.
  • Additional CLI notes and help command reference.
  • Documented AGI_INSIGHT_MEMORY_PATH for persistent storage configuration.
  • Optional CPU and memory caps for the CodeGen sandbox via SANDBOX_CPU_SEC and SANDBOX_MEM_MB.
  • Optional firejail sandboxing when the binary is available.
  • Configurable secret backend for HashiCorp Vault and cloud managers via AGI_INSIGHT_SECRET_BACKEND.
  • Protobuf dataclasses generated via make proto (uses the betterproto plugin when installed).
  • Optional JSON console logging and DuckDB ledger support.
  • Aggregated forecast endpoint /insight and OpenAPI schema exposure.
  • Baseline load test metrics: p95 latency below 180 ms with 20 VUs.
  • Prometheus metrics available at /metrics.
  • Lightweight /status endpoint listing agent names, heartbeats and restarts.
  • GraphMemory._fallback_query now returns stub data when both Neo4j and NetworkX are missing.
  • Property-based tests verify SafetyGuardianAgent blocks malicious code.
  • Added optional e2e test broadcasting Merkle roots to the Solana devnet.
  • Example Helm values file for the Insight chart.
  • simulate command now accepts --energy and --entropy to set initial sector values. The React dashboard exposes matching input fields.
  • Version constant __version__ defined in alpha_factory_v1.demos.alpha_agi_insight_v1.__init__.
  • Version constant __version__ defined in alpha_factory_v1.demos.cross_industry_alpha_factory.__init__.

[v1.0] - 2025-07-01

  • CLI commands simulate, show-results, agents-status and replay for running and inspecting forecasts.
  • Web dashboard served when RUN_MODE=web or via streamlit run src/interface/web_app.py.
  • Security features include API token authentication, optional TLS for the gRPC bus and REST API, ED25519-signed agent wheels and HashiCorp Vault integration for secrets.

[1.0.2] - 2025-06-30

  • Documented CLI and FastAPI usage with example commands.
  • Added agent architecture diagrams and algorithm explanations for MATS and forecasting.
  • Provided a Streamlit dashboard for interactive demos.
  • Maintained versioned changelog for project history.

[1.0.1] - 2025-05-25

  • Packaged the React web client and served static assets from the API server.
  • Added asynchronous API tests and new CLI flags.
  • Improved container orchestration scripts and default Docker compose files.

[1.0.0] - 2025-05-15

  • Initial release with an offline-friendly CLI and REST API for running simulations.
  • Included a minimal web interface served when RUN_MODE=web.
  • Shipped Meta-Agentic Tree Search and forecasting modules with Docker deployment scripts.

1.5.0 — Ascension and the interactive white paper

Adds a connected six-stage lab with three editable flagship scenarios, real portfolio/schedule algorithms, authenticated Nova-Seed encryption/recovery, exact modeled token accounting and policy search. The governance observatory exposes replicator dynamics, risk corrections, stake/slash controls, quadratic credits and an eight-day gate. Redesigns the Pages entry point while preserving every original demo and flywheel.

See release notes and implementation evidence.

1.17.0 — MATS Search Lab

See release notes for bounded branching search, exact browser/native replay, independent gates, preserved research and publication acceptance.