ab9d1c0fa4
Some checks failed
Smoke Test / smoke (pull_request) Failing after 23s
Architecture Lint / Linter Tests (pull_request) Successful in 26s
Validate Config / YAML Lint (pull_request) Failing after 15s
Validate Config / JSON Validate (pull_request) Successful in 19s
Validate Config / Python Syntax & Import Check (pull_request) Failing after 1m1s
Validate Config / Python Test Suite (pull_request) Has been skipped
Validate Config / Shell Script Lint (pull_request) Failing after 1m4s
Validate Config / Cron Syntax Check (pull_request) Successful in 13s
Validate Config / Deploy Script Dry Run (pull_request) Successful in 13s
Validate Config / Playbook Schema Validation (pull_request) Successful in 25s
Architecture Lint / Lint Repository (pull_request) Failing after 22s
PR Checklist / pr-checklist (pull_request) Successful in 5m0s
Add fleet.inventory and fleet.path_contracts to config.yaml: - Central source of truth for IPs, ports, roles, remote paths - Introduce get_config_path(), load_fleet_inventory(), get_path_contract() - Updated fleet_llama.py, self_healing.py, telemetry.py, agent_dispatch.py, skill_installer.py to read from config instead of hard-coded dicts/paths - Documented inventory contract and override mechanism in scripts/README.md Scripts retain forward-compatible fallback defaults for backwards compatibility. Closes #433
117 lines
4.9 KiB
Markdown
117 lines
4.9 KiB
Markdown
# Gemini Sovereign Infrastructure Suite
|
|
|
|
This directory contains the core systems of the Gemini Sovereign Infrastructure, designed to systematize fleet operations, governance, and architectural integrity.
|
|
|
|
## Principles
|
|
|
|
1. **Systems, not Scripts**: We build frameworks that solve classes of problems, not one-off fixes.
|
|
2. **Sovereignty First**: All tools are designed to run locally or on owned VPSes. No cloud dependencies.
|
|
3. **Von Neumann as Code**: Infrastructure should be self-replicating and automated.
|
|
4. **Continuous Governance**: Quality is enforced by code (linters, gates), not just checklists.
|
|
|
|
## Tools
|
|
|
|
### [OPS] Provisioning & Fleet Management
|
|
|
|
- **`provision_wizard.py`**: Automates the creation of a new Wizard node from zero.
|
|
- Creates DigitalOcean droplet.
|
|
- Installs and builds `llama.cpp`.
|
|
- Downloads GGUF models.
|
|
- Sets up `systemd` services and health checks.
|
|
- **`fleet_llama.py`**: Unified management of `llama-server` instances across the fleet.
|
|
- `status`: Real-time health and model monitoring.
|
|
- `restart`: Remote service restart via SSH.
|
|
- `swap`: Hot-swapping GGUF models on remote nodes.
|
|
- **`skill_installer.py`**: Packages and deploys Hermes skills to remote wizards.
|
|
- **`model_eval.py`**: Benchmarks GGUF models for speed and quality before deployment.
|
|
- **`phase_tracker.py`**: Tracks the fleet's progress through the Paperclips-inspired evolution arc.
|
|
- **`cross_repo_test.py`**: Verifies the fleet works as a system by running tests across all core repositories.
|
|
- **`self_healing.py`**: Auto-detects and fixes common failures across the fleet.
|
|
- **`agent_dispatch.py`**: Unified framework for tasking agents across the fleet.
|
|
- **`telemetry.py`**: Operational visibility without cloud dependencies.
|
|
- **`gitea_webhook_handler.py`**: Handles real-time events from Gitea to coordinate fleet actions.
|
|
|
|
### [ARCH] Governance & Architecture
|
|
|
|
- **`architecture_linter_v2.py`**: Automated enforcement of architectural boundaries.
|
|
- Enforces sidecar boundaries (no sovereign code in `hermes-agent`).
|
|
- Prevents hardcoded IPs and committed secrets.
|
|
- Ensures `SOUL.md` and `README.md` standards.
|
|
- **`adr_manager.py`**: Streamlines the creation and tracking of Architecture Decision Records.
|
|
- `new`: Scaffolds a new ADR from a template.
|
|
- `list`: Provides a chronological view of architectural evolution.
|
|
|
|
## Usage
|
|
|
|
Most tools require `DIGITALOCEAN_TOKEN` and SSH access to the fleet.
|
|
|
|
```bash
|
|
# Provision a new node
|
|
python3 scripts/provision_wizard.py --name fenrir --model qwen2.5-coder-7b
|
|
|
|
# Check fleet status
|
|
python3 scripts/fleet_llama.py status
|
|
|
|
# Audit architectural integrity
|
|
python3 scripts/architecture_linter_v2.py
|
|
```
|
|
|
|
|
|
## Fleet Inventory Contract
|
|
|
|
The fleet inventory is defined in `timmy-config/config.yaml` under the `fleet:` key. All [OPS] scripts read this data at runtime, eliminating hard-coded IPs and paths.
|
|
|
|
### `fleet.inventory` — Per-Host Definition
|
|
|
|
```yaml
|
|
fleet:
|
|
inventory:
|
|
<hostname>:
|
|
ip: <string> # Public or private IP address
|
|
port: <int> # SSH target port (typically 22)
|
|
role: <string> # Logical role (hub, forge, agent-host, world-host)
|
|
remote_root: <path> # Remote root directory for Hermes operations
|
|
capabilities: [...] # Feature tags the host supports
|
|
```
|
|
|
|
Each host entry exposes: `ip`, `port`, `role`, `remote_root`, `capabilities`. The `capabilities` tag is freeform but standardized across the fleet (e.g., `gateway`, `orchestrator`, `forge`, `agent-host`, `llm-host`, `world-host`).
|
|
|
|
### `fleet.path_contracts` — Path Abstractions
|
|
|
|
```yaml
|
|
fleet:
|
|
path_contracts:
|
|
hermes_agent_local: ../hermes-agent # Path to local hermes-agent repo (relative to timmy-config)
|
|
hermes_remote: /opt/hermes # Remote Hermes root on fleet nodes
|
|
skills_remote: /opt/hermes/skills # Remote skills directory
|
|
```
|
|
|
|
All scripts reference paths via `get_path_contract(key, default)` or `get_remote_root()` helpers. This centralizes path management across local (mac) and remote wizards.
|
|
|
|
### Override Mechanism
|
|
|
|
Set the `TIMMY_CONFIG` environment variable to point at an alternate `config.yaml`:
|
|
|
|
```bash
|
|
export TIMMY_CONFIG=/path/to/alternate/config.yaml
|
|
python3 scripts/fleet_llama.py status
|
|
```
|
|
|
|
Without `TIMMY_CONFIG`, scripts auto-resolve `timmy-config/config.yaml` relative to their `scripts/` directory.
|
|
|
|
### Fallback Defaults
|
|
|
|
If `config.yaml` is missing or the `fleet:` section is absent, scripts fall back to the canonical production fleet:
|
|
|
|
| Hostname | IP | Role |
|
|
|----------|-------------------|---------------|
|
|
| mac | 10.1.10.77 | hub |
|
|
| ezra | 143.198.27.163 | forge |
|
|
| allegro | 167.99.126.228 | agent-host |
|
|
| bezalel | 159.203.146.185 | world-host |
|
|
|
|
Fleet eviction occurs through config changes, not code edits.
|
|
|
|
---
|
|
*Built by Gemini — The Builder, The Systematizer, The Force Multiplier.*
|