Promethea Agent v0.1.0 Public Preview

Release Notes

Public Preview Release Candidate

This release candidate focuses on making Promethea usable as a local-first cognitive agent runtime with a modern Web UI, graph-first memory, backend-aware first-run behavior, and release-ready documentation.

Highlights

• Promotional bilingual homepage for Promethea's positioning: beyond memory, toward cognition.
• Vite Web UI connected to the gateway startup script.
• Backend-aware first-run auth flow:
  • Neo4j remains the default and core graph backend.
  • If Neo4j is configured but unavailable, registration is blocked with a clear message instead of silently falling back.
  • The login screen can enter setup/diagnostics mode before sign-in.
• File upload and attachment flow:
  • Text-like files can be extracted and attached to chat.
  • Image files are stored and may use OCR when optional dependencies are available.
  • Stored-only attachments are explicitly described as unavailable to the model instead of being invented.
• Global search across sessions and uploaded files.
• Workflow inspector for definitions, personal runs, recovery, pause/resume, and checkpoints.
• Enterprise Brain visibility now follows org_brain.enabled; disabled enterprise features are hidden from the UI.
• Memory visualization and inspector readability improvements.
• Test artifact policy confines runtime test files under .tmp/pytest-runtime/.
• User configuration is split into sensitive env values, basic config, and advanced config.
• Prompt assembly uses explicit prompt blocks, with policy routing prepared for need-based block selection.
• memory/raw_log.jsonl is documented as the L0 memory layer, not disposable test output.

Important Behavior

• Full graph-memory behavior requires Neo4j.
• Fallback memory backends remain available, but they are explicit choices, not silent replacements for the default Neo4j experience.
• Changing backend configuration should be followed by a restart.
• Playwright browser control requires installing Playwright browsers separately.
• Multimodal support currently means file storage, text extraction, and optional OCR; it is not guaranteed native vision input for every model/provider.
• Voice input is not part of the supported preview release surface. The current experimental push-to-talk route depends on an OpenAI-compatible audio transcription provider; a DeepSeek-only chat configuration does not provide STT/audio transcription.

Verification

Release-focused checks run during preparation:

• python -m pytest: 500 passed, 5 skipped
• npm run build in UI/: TypeScript and Vite production build passed
• npm run lint in UI/: ESLint passed

Manual checks still required before a stable tag:

• Web UI smoke test: first-run status, registration/login, chat, file attachment, global search, memory inspector, workflow inspector, settings save, enterprise feature visibility.
• Neo4j-on and Neo4j-off first-run flows.

Known Limitations

• This is a public preview / release candidate, not a final stable commercial release.
• Neo4j is strongly recommended for the intended graph-structured memory experience.
• Some advanced integrations depend on optional local services or third-party credentials.
• Frontend development/build can be affected by local Windows Node/esbuild execution policy; verify in a normal local PowerShell environment.
• Voice endpoints may appear in the internal API surface, but they are experimental/provider-dependent and should not be presented as a ready user feature in this release.

Upgrade Notes

• Copy new environment variables from env.example if your local .env is old.
• Restart the gateway after changing memory backend or enterprise feature switches.
• Do not commit files under config/users/, logs/, workspace/, UI/node_modules/, .tmp/, or local test artifact directories.
• Treat memory/raw_log.jsonl and memory/raw_log.state.json as runtime memory state. Do not clean them as generic logs.