Skip to content
Agents tracked: 501 Releases (7d): 777 Agent status: no reported issuesSince last limit reset: Claude 10d · Codex 8d For AI agents: Markdown Updated Nov 21, 2026

ARTEX changelog: what's new each month

Every stable ARTEX release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 2 months covered; the current month updates daily.

November 2026 · so far

3 releases: 0.5.0 → 0.6.0

November brought Ollama support for local language models, a fresh security overhaul with setup tokens and secrets isolation, and significant security hardening including verified findings with proof-of-evidence levels and target gating for approvals.

Highlights

  • Added Ollama support to run language models locally or on a LAN without requiring API keys.
  • Implemented setup tokens for secure first-run configuration, preventing unauthorized access to fresh installations.
  • Introduced verified findings with proof-of-evidence levels and verification status tracking.
  • Added target gating to control and audit approval decisions with IP resolution.
  • Hardened the agent with code-owned rules, goal guards, and improved audit logging.

New

  • Ollama LLM profile format for local or LAN servers with automatic base URL normalization to /v1 root.
  • Setup token system: random one-time token generated at startup and logged for first-run configuration.
  • Verified findings feature tracking proof-of-evidence levels and verification status with bound evidence requirements.
  • Target gate functionality for approvals that pins host:port and resolved IPs with opt-in per task or conversation.
  • Intent claim leases with heartbeat, TTL and owner token to prevent stale workers from overwriting results.
  • Configurable notification limits for summaries and assets with Feishu digest overflow handling.

Improved

  • ARTEX_BIND_ADDR configuration for controlling which address the web UI publishes on.
  • ARTEX_LLM_PROVIDER now matched case-insensitively to prevent fallback to Anthropic defaults.
  • API keys now scoped to their own endpoint, preventing reuse across different base URLs.
  • Startup secrets removed from environment after reading to reduce exposure window.
  • Docker supply chain verification including recon tool checksums, sqlmap pinning, and Node.js verification.
  • Intercept audit now records decided_by field and fixes model prefix bug in judge decision storage.
  • Custom User-Agent header support per LLM profile.
  • List task credentials schema corrected.

Fixed

  • Ollama servers configured via ARTEX_LLM_BASE_URL no longer return 404 errors on chat endpoints due to missing /v1 path normalization.
  • Saved API keys were previously being used for any base_url in test requests; now scoped to their own endpoints.
  • ARTEX_LLM_PROVIDER case-sensitivity issue where values like Ollama fell through to Anthropic defaults.
  • First-run claim vulnerability allowing anyone with port access to reach the web UI before admin password setup.
  • Custom tool proxy scheme configuration issue.
  • Model prefix bug in audit logging for judge decisions.

October 2026

1 release: 0.4.0

October brought a major expansion of ARTEX's reconnaissance and exploitation capabilities, with integrated security tools, policy-aware scanning skills, and improved findings validation. The release also strengthened operational safety and cost tracking for sustained agent autonomy.

Highlights

  • Recon toolbox now built into the image with nuclei, httpx, subfinder, katana, ffuf, dalfox, and sqlmap at pinned versions.
  • Recursive recon skill uses bbot v3.0.2 to run passive scans when programs ban active scanning, switching to active modules only when permitted.
  • New php2shell skill upgrades file-inclusion findings to verified RCE with policy gating and lab testing first.
  • Findings now validated through proof-basis gates, measured impact assessment, adversarial verification, and deduplication rules.
  • Per-task credentials vault, token-level cost ledger, and stagnation warnings help prevent resource waste and repetitive scanning.

New

  • Recursive recon skill with policy-aware passive and active module switching.
  • php2shell skill to verify file-inclusion vulnerabilities as working RCE.
  • Goal ambiguity gate at task creation to clarify intent.
  • Stagnation warning in graph overview when workers repeat without new findings.
  • Per-task credentials vault for isolated secret management.
  • Cost ledger with per-call token accounting.
  • LLM chain failure escalation and per-intent profile rotation.
  • MCP health reporting.

Improved

  • Findings discipline system with proof-basis gates and adversarial verification.
  • Norma upgraded to v0.4.4 with per-task process-tree memory watchdog.
  • UTF-8 aligned truncation for safer message handling.
  • Non-interactive subprocess environments to prevent unintended prompts.
  • message_delta usage recovery for more efficient token tracking.
  • Streaming fixes for more reliable output handling.

Fixed

  • Streaming issues resolved in Norma v0.4.4.

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. ARTEX: pricing, features and alternatives · All changelogs