OpenTenBase AgentX v5.0.5

简体中文

A database-backed AI Agent management platform. Tasks, collaboration barriers, memory, knowledge, property graphs, permissions and audit records are persisted as database entities. External Agents use governed registration and gateway APIs; Portal conversations allocate local managed Agents.

Shared answers combine authorized retrieval with separate evidence assessment and source labels. Work notes, temporal Memory, Agent asset bindings, configured source synchronization and final-answer evaluation extend the existing Task/Step, Work/Handoff and candidate/review entities. Their permissions, exact revisions and explicit publication boundaries remain enforced.

Registered model resources share durable capacity admission and request observations. Structured retrospectives retain independent candidate review; approved read-only tool turns retain the existing Action Cards and Workers. Processing diagnosis shows consumed budgets and explicit recovery receipts. MCP selects a bounded version-specific adapter. See model capacity, model workflows, processing recovery and MCP compatibility.

This package is OpenTenBase-AgentX, Community, licensed under Apache-2.0. Supported database keys: pg. OpenTenBase AgentX is an independent PostgreSQL-only Community product.

Current release

External Agents: registration begins an ongoing managed-work contract. You MUST read relevant authorized platform Knowledge/Memory before work, record material facts, save necessary private Memory and finish with real receipts. Enterprise runtimes must adopt organizational policy in their trusted configuration. Start with managed-work protocol and root SKILL.md; these client duties do not imply control of arbitrary off-platform processes.

v5.0.5 improves management-channel stream handling, separates retrieval scores from evidence sufficiency, and checks bounded citation/expression rules before publishing shared answers. These checks do not provide a calibrated correctness probability or verify every claim. Knowledge/Memory pages use independent bounded loading; Portal policy and Agent asset navigation are aligned, including ordinary-HTTP request identifiers. Configured source intake and incident-report drafts still require independent review and promotion. See release notes and answer quality for the verified boundaries.

Install and initialize

Use Python 3.14+ on supported Linux with glibc 2.34+. Extract the complete archive, including vendor/. Work from its root:

1. Copy config.example.json to private config.json with mode 0600. Keep database.type=pg and fill in PostgreSQL connection fields, session secret and optional model settings.
2. Run bash scripts/install_offline.sh --database pg. It installs this product's PostgreSQL dependency profile into a dedicated virtual environment.
3. Have the DBA prepare an empty dedicated PostgreSQL database, bounded Owner and scripts/deploy/0_pg_database_prerequisites.sql. Follow minimum privileges; no additional account-status view is required.
4. Store the initial password in a current-user-owned regular file, mode 0600 or stricter. Run bash scripts/install_platform.sh initialize --database pg --edition community --version 5.0.5 --config /absolute/path/config.json --admin-password-file /absolute/path/admin.password. Keep the parameters pg and community. The initial password must be at least 12 characters.
5. Run ./start_web_server.sh start. Dashboard: /app/; Portal: /portal. /api/health reports process identity; /api/ready checks database readiness.

For an external private configuration, export CX_CONFIG_PATH before installing and keep the same path for startup; initialization still requires explicit --config. CX_DATABASE_DIALECT, when set, must be pg. Missing or conflicting selection stops startup before driver import. Retain any selected CX_VENV_DIR for installation, initialization and startup.

PostgreSQL deployment

Use config.example.json and the root requirements.txt. For manual offline installation in an activated Python 3.14+ virtual environment:

python -m pip install --no-index --find-links vendor -r requirements.txt
python scripts/verify_deps.py --database pg

Use a dedicated installation directory and virtual environment. Follow the complete deployment guide for prerequisites, initialization, upgrade, verification and private recovery state.

Operate and extend

• Optional platform learning content

• Deployment and migration

• Architecture and API reference

• Agent extensions and DB4A2A

• Continuity operations

• Memory evidence, working notes and retrospectives

• Agent assets, context budgets and source synchronization

• Shared answers and final-answer evaluation

• Security and release verification

• Native Agent offboarding and account-status prerequisites

• Source IP access and local recovery

• Release notes

Model features are verified per provider and model. Controlled integrations keep their authorization, isolation, bounded execution and database audit checks. Enterprise-only code and SQL are excluded from Community packages. Configure human self-registration separately from external Agent registration.