Ares

Ares is an operator-supervised security assessment runtime for authorized engagements. It gives a model room to reason and request actions while keeping scope, risk, approvals, routing, persistence, evidence, and reporting under deterministic operator control.

Ares is designed for security practitioners who need more than a chat wrapper but do not want an unattended agent making arbitrary offensive decisions.

Authorized testing only. Do not use Ares against systems you do not own or do not have explicit permission to assess.

Product identity

• Product: Ares
• Python distribution: bluedot-ares
• Python import: ares
• Primary command: ares
• Current source version: 1.1.1
• License: MIT
• Supported Python: 3.11, 3.12, and 3.13

The PyPI project named ares belongs to an unrelated package. Do not install it expecting this project. Install bluedot-ares in an isolated environment.

What Ares provides

Ares combines four execution paths behind the same policy and evidence model:

The runtime includes:

• OpenAI-compatible, Anthropic, and Gemini model adapters
• provider fallback chains and OAuth helpers
• a central tool registry with availability, schema, risk, and toolset metadata
• dispatcher-owned scope, rules of engagement, risk, approval, duplicate, route, and timeout enforcement
• governed autonomous reconnaissance with a persistent attack-surface graph and coverage ledger
• a finding lifecycle from observation through corroboration, safe validation, and reporting
• evidence-bound, single-use approval receipts for advanced operator validation
• SQLite state for sessions, messages, tool calls, hosts, services, missions, findings, and memory
• GhostMCP and bounded OnionClaw integrations
• Markdown session and mission reports
• separate gateway, dashboard, and terminal operator surfaces

Install

The recommended end-user installation is an isolated application environment:

pipx install bluedot-ares
# or
uv tool install bluedot-ares

If the matching bluedot-ares release is not available from PyPI, install its verified wheel from the GitHub release or use the source installation path in INSTALL.md.

Verify the install:

ares --version
ares doctor
ares doctor --json

Optional provider and tool integrations remain extras:

pipx install 'bluedot-ares[anthropic]'
pipx install 'bluedot-ares[gemini]'
pipx install 'bluedot-ares[ghostmcp]'

See INSTALL.md for upgrades, source installs, Windows notes, and migration from the older GitHub wheel.

Five-minute start

Run the guided setup:

ares onboard

For a local OpenAI-compatible model server:

export LLM_PROVIDER="local"
export LLM_MODEL="local-model"
export OPENAI_BASE_URL="http://127.0.0.1:1234/v1"
export OPENAI_API_KEY="lm-studio"

Check the selected route without executing tools:

ares route \
  --target 127.0.0.1 \
  --prompt "Enumerate the authorized loopback target and stop after useful initial findings."

Run a bounded loopback assessment:

ares run \
  --target 127.0.0.1 \
  --prompt "Enumerate the authorized loopback target and stop after useful initial findings." \
  --max-iterations 20

Higher-risk actions are denied by default. --approve-dangerous satisfies only the dispatcher approval gate. Scope, risk, route, and tool policy still run before execution.

The full walkthrough is in docs/quickstart.md.

Governed autonomous reconnaissance

The autonomous-recon mission profile allows a model to prioritize reconnaissance work without giving it arbitrary tool or target construction authority.

ares mission run \
  --profile autonomous-recon \
  --target 192.0.2.10 \
  --allowed-host 192.0.2.10 \
  --max-risk active \
  --ports 22,80,443,8000 \
  --max-tasks 16 \
  --autonomous \
  --out autonomous-recon.md

Ares persists each attack-surface node, relationship, coverage item, planner decision, tool result, recovery attempt, finding hypothesis, and limitation. The planner can choose only from exact coverage IDs compiled by Ares into fixed tools, targets, and arguments.

This path performs passive and safe-active reconnaissance. It does not perform exploitation, authentication attempts, persistence, or arbitrary command execution.

Authorized operator validation

Advanced validation uses an explicit operator-supplied task graph. It requires same-mission evidence, a GhostMCP engagement policy where applicable, an immutable approval receipt bound to the exact task digest, and an out-of-model approval.

ares mission run \
  --profile authorized-operator-validation \
  --mission-id engagement-2026-001 \
  --target 192.0.2.10 \
  --allowed-host 192.0.2.10 \
  --max-risk post-exploitation \
  --initial-tasks tasks.json \
  --ghostmcp-policy engagement-policy.json \
  --approval-receipts approval-receipts.json \
  --approve-high-risk

The model never selects advanced-role tasks. Failed, empty, cross-target, cross-mission, expired, replayed, or contract-mismatched evidence and receipts fail closed.

Operator surfaces

Gateway

The gateway is the backend API and control plane for authentication, pairing, allowlists, run submission, run status, event polling, and audit logging.

ares gateway-config --mode loopback
ares gateway

Remote exposure should use bearer authentication and a CIDR allowlist:

ares gateway-config \
  --mode exposed \
  --auth-enabled \
  --allow-cidr 203.0.113.0/24

Dashboard

The dashboard is the browser frontend backed by the gateway:

ares dashboard
# or
ares-dashboard

For a remote server:

ares dashboard --mode lan --no-open

Terminal UI

ares tui
# or
ares-tui

The gateway, dashboard, and TUI are separate product surfaces even when the gateway serves bundled dashboard assets.

Diagnostics and support

Machine-readable diagnostics:

ares doctor --json

Create a redacted support bundle that excludes credentials and engagement evidence:

ares support-bundle --out ares-support-bundle.json

Read SUPPORT.md before opening an issue. Installation and runtime failures are covered in docs/troubleshooting.md.

Offline Evaluation Lab

Run the bundled deterministic regression fixtures without a model, network request, or engagement data:

ares evaluate
ares evaluate --out ares-evaluation.json

The result measures exact agreement with the versioned fixtures for mission and task outcomes, scope and policy decisions, and finding-validation readiness. It does not measure model quality, live-target behavior, vulnerability discovery, or broad security efficacy. See docs/evaluation-lab.md for metric semantics and limitations.

Release verification

Official releases include:

• wheel and source distribution
• SHA256SUMS
• CycloneDX SBOM
• release metadata
• GitHub build provenance and SBOM attestations

Verification commands are documented in docs/verifying-releases.md.

Architecture

The model does not call tools directly. Ares converts a model request or mission decision into a dispatcher request, validates it against policy and authorization state, records the result, indexes evidence, and then renders operator-visible state and reports.

See docs/architecture.md and docs/v1-support-boundary.md for the supported contract.

Development

git clone https://github.com/BlueDot-IT/Ares.git
cd Ares
git submodule update --init --recursive

python -m venv .venv
. .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e '.[dev,anthropic,gemini,ghostmcp]' -e vendor/ghostmcp

python -m pytest tests -q
python -m compileall src/ares
python -m build
python -m twine check dist/*

Contribution requirements are in CONTRIBUTING.md.

Documentation

• Installation
• Quickstart
• Mission CLI
• Evaluation Lab
• Architecture
• Troubleshooting
• Release verification
• Support boundary
• Security policy
• Support policy
• Changelog

License

Ares is released under the MIT License.