Haephestus
A sovereign software and AI lifecycle platform for source, collaboration, artifacts, computational models, repository intelligence, agent authority, builds, verification, delivery and repair.
Haephestus
A sovereign software and AI lifecycle platform for source, collaboration, artifacts, computational models, repository intelligence, agent authority, builds, verification, delivery and repair.
One Platform, Not a Toolchain
Source control, container registry, ML tracking, code search, and deployment — one platform, one identity, one protocol, one audit trail. No bolting together disconnected source, artifact, and model systems with separate auth.
Built in Rust
No Go services, no Python sidecars, no Node.js microservices. One Rust binary with native Git engine, native OCI registry, and native ML tracking — fast, memory-safe, and deployable anywhere.
Defense-Grade Deployment
Seven deployment classes including air-gapped for classified networks and edge-constrained for industrial sites. Quantum-resistant signing. Most platforms stop before this operational range begins.
ML and Code, Same Pipeline
Models use the same registry, same identity, same deployment gates as containers. No separate ML platform handoff — model promotion uses the same approval chain as a production deploy.
Sovereign, No Upstream Dependency
No dependency on a patchwork of upstream release schedules or licensing changes. Haephestus owns the entire surface — every line of code, every protocol, every security decision.
Full Supply-Chain Provenance
Every commit, every artifact, every model version, every deployment — traceable with deterministic provenance and tamper-evident audit trails. SBOM, vulnerability scanning, and compliance evidence built in.
Most engineering teams run a Frankenstein toolchain — one service for source code, another for containers, a third for experiments, a separate build controller, and a wiki nobody updates. Each has its own login, its own permissions, its own audit trail, and its own bill. When a security audit asks who pushed what image to production and which model version is deployed, the answer requires correlating logs across four systems that were never designed to talk to each other.
Haephestus is one platform that handles the entire software supply chain — source control, code review, CI/CD, container registry, package registry, ML experiment tracking, model registry, code search, and deployment — under one identity system, one permission model, one protocol surface, and one audit trail. Built in Rust. No Go services. No Python sidecars. No separate auth for the registry that the security team set up without telling anyone.
Haephestus now carries the lifecycle beyond repository and registry operations: semantic code assurance, authorized repository intelligence, bounded agent mutation, change-impact landing gates, remote build and verification, blast-radius-aware delivery, and artifact anti-entropy all share the same identity, evidence and custody spine.
The complete Haephestus platform
Source Control and Collaboration
Host repositories, review changes and coordinate software work under one sovereign identity and evidence model.
Git Engine and Code Search
Operate repository storage, history and search as native Haephestus services rather than opaque external islands.
Artifact and Package Registry
Store, verify, replicate and distribute containers, packages, models, firmware and generic artifacts as governed release inputs.
AI, ML and Computational-Model Lifecycle
Govern datasets, feature releases, experiments, models, checkpoints, evaluations and composed releases across learned and non-learned computation.
Identity, Policy and Sovereign Runtime
Apply organization, tenant, identity, policy and deployment controls consistently across source, artifacts, agents and delivery.
Developer and Operator Experience
Expose coherent GUI, CLI, SDK, API, observability and support surfaces over the same product contracts.
Semantic Code Assurance
Evaluate code structure, behavior, architecture and policy across exact source revisions before a change lands.
Repository Intelligence and RAG
Turn authorized source, specifications, graphs and evidence into explainable repository understanding.
Agent Authority and Change Landing
Grant bounded agent mutation, prove change impact and land protected changes only against current evidence.
Build and Continuous Verification Fabric
Accelerate builds and tests while preserving hermetic inputs, trust, reproducibility and complete verification debt.
Continuous Delivery and Blast-Radius Orchestration
Move accepted changes through evidence-bound stages, cohorts, observation and rollback across many deployable units.
Artifact Custody and Release Integrity
Keep every software, model, firmware and deployment artifact available, repairable and attributable through its entire lifecycle.
What Haephestus Replaces
Every engineering team runs a toolchain that was assembled over years — someone needed a Git host, someone else needed a container registry, the ML team set up their own tracking, and security bolted on after the fact. Each tool has its own login, its own permissions, its own API, its own audit trail, and its own bill. Haephestus replaces all of them with one platform.
Repository hosting, code review, pull requests, issues, wikis, CI/CD pipelines, webhooks, releases, migrations, mirrors — the complete collaborative development surface, built in Rust with one identity system and one audit trail. Your developers use the same Git workflows they already know.
The Git engine is native Rust — not a wrapper around the C Git binary. Object storage, pack files, refs, diff, merge, rebase, transport, LFS, submodules — all native. No shelling out to external processes. Faster, safer, and debuggable as one system.
One Identity, One Protocol
The biggest security risk in a toolchain is not any single tool — it is the gaps between them. Different tools have different permission models, different audit formats, and different ways of answering 'who did what, when, and why.' Haephestus eliminates the gaps by giving every vertical the same identity, the same protocol, and the same audit trail.
One login. Every tool. The developer who pushes code, the DevOps engineer who deploys the image, and the data scientist who promotes the model are the same person in the same system — not three identities across three vendor consoles with three different permission models.
When a new developer joins, you grant access once — not once per tool. When a contractor leaves, you revoke once — not hope you remembered the detached registry account they set up on a Tuesday. When a security audit asks who has access to push production images, the answer is one query — not a spreadsheet you maintain by hand.
Deployment Freedom
Most DevOps platforms assume you run in the cloud. Haephestus runs anywhere — from a developer laptop to a classified air-gapped network to an industrial edge device. Seven deployment classes, all from the same binary, all with the same protocol surface.
Run Haephestus on your own infrastructure — bare metal, VM, or Kubernetes. Your code, your containers, your models, your data — all on your network. No outbound telemetry. No phone-home. The same platform that runs in your data center is the same platform that runs in the cloud.
Who Uses Haephestus
Data Scientist
Trains models with full experiment provenance — training data, hyperparameters, code version, validation metrics. Promotes models through the same registry as containers. No separate ML platform with its own auth and its own audit trail.
DevOps Engineer
Deploys containers and models through the same gates — identity, permissions, scanning, signing. One CLI, one API, one audit trail for both. No context switching between GitHub, Harbor, and MLflow.
Security Auditor
Traces any artifact from commit to deployment in one query. SBOM, vulnerability history, and approval chain — all in one tamper-evident log. No cross-referencing across four vendor consoles.
Platform Owner
Manages one platform instead of four. One deployment, one upgrade cycle, one monitoring dashboard, one bill. Air-gapped, edge, or cloud — the same binary, the same governance.
ML to Production, One Pipeline
The gap between ML and production is the most dangerous gap in your toolchain. A data scientist trains a model in MLflow. A DevOps engineer deploys a container in Harbor. Nobody checks whether the model in production matches the model that was validated. Haephestus closes this gap — models and containers go through the same registry, the same approval chain, and the same deployment gates.
Model versioning with the same rigor as container images — signed, scanned, provenance-tracked. When a model is promoted from staging to production, it goes through an approval chain with evidence: training data, validation results, bias checks, performance metrics. No model reaches production without a documented, auditable approval.
UI and Developer Experience
One web interface for source control, registry, ML tracking, and deployment. Not a portal that links to separate tools — one application with one navigation, one design language, and one identity. Every action in the UI is backed by the same protocol operation as the CLI and the API.
One dashboard for the entire software supply chain — repositories, registries, ML experiments, deployments, and security. Deployment-class-aware: the UI adapts to your deployment mode — full in self-hosted, governed in managed, bounded in edge, reduced in air-gapped. No features you cannot use, no missing features you need.
Tool Consolidation: Before vs After
One platform replaces 6+ tools. Fewer logins, fewer bills, fewer security gaps. GitHub, Harbor, MLflow, Jenkins, code search, and wiki — all consolidated into one system with one identity and one audit trail.
Your security team reviews one attack surface, not six. Your finance team pays one invoice, not six. Your engineers learn one API, not six.
Deployment Class Distribution
Seven deployment classes — one binary serves all, from dev laptop to classified network. Self-hosted on your metal, managed in our cloud, bring-your-own-cloud, hybrid, air-gapped, edge, or shared SaaS.
Same code. Same API. Same audit trail. Different deployment boundaries for different security postures.
Code intelligence that can prove its answer
A repository question begins with an authorized immutable revision. Haephestus retrieves exact source and graph evidence, labels inference, exposes omissions and keeps the generation identity beside every answer.
Source
Immutable repository and specification evidence remains authoritative.
Graph
Symbols, calls, control flow, data flow, routes and generated lineage explain relationships.
Synthesis
Model output remains bounded, attributable and never becomes source truth.
Agents work inside grants, not around them
An agent receives an exact repository, revision, branch, path, operation, tool, provider, time and evidence envelope. Every diff, commit, push and pull request remains attributable; approval, protected landing, release and deployment stay separately authorized.
Build speed never outranks trust
Remote execution, caches and warm workspaces accelerate heterogeneous builds only when toolchain, inputs, sandbox, tenant, producer and reproducibility dimensions match. Unknown dimensions force verification or a miss.
Delivery follows the actual blast radius
Every deployable unit receives a relevant-commit history, current impact snapshot, required verification and stage-gate evidence. Cohorts advance on positive evidence, pause on negative evidence and treat missing evidence as unresolved—not success.
Use-Case Surfaces
- Replace Your ToolchainOne platform replaces GitHub, Harbor, MLflow, and code search — one identity, one protocol, one audit trail.
- Air-Gapped DevOpsRun source control, registry, CI/CD, and ML tracking in classified networks — no internet, no phone-home, quantum-resistant signing.
- ML to ProductionModels and containers go through the same registry, same approval chain, same deployment gates — no gap between ML and DevOps.
- Supply-Chain SecuritySBOM, vulnerability scanning, provenance, and tamper-evident audit for every artifact, model, and deployment.
- Edge DevOpsRun CI/CD and registry on factory floors, ships, and remote sites with intermittent connectivity and bounded resources.
- Sovereign InfrastructureNo dependency on upstream GitHub, GitLab, Harbor, or MLflow release schedules. Haephestus owns the entire surface.
Your Toolchain Is a Security Risk
Your source control is in GitHub. Your container registry is in Harbor. Your ML tracking is in MLflow. Your CI is in Jenkins. Each has its own login, its own permissions, its own API, and its own audit trail. When a security auditor asks who pushed a specific image to production and which model version is running, the answer requires correlating logs across four systems that were never designed to talk to each other. The gaps between your tools are where security incidents hide.
Haephestus is one platform. Source control, registry, ML tracking, code search, and deployment share one identity, one protocol, one audit trail, and one deployment model. The person who pushes code, the person who pushes an image, and the person who promotes a model are the same identity with the same permissions under the same policy. When an auditor asks a question, the answer comes from one system — not from a cross-referencing exercise across four vendor consoles.
That is what Haephestus is. Not another tool to add to your toolchain. A replacement for the toolchain itself.
Proof points
- 4 verticals in one platform with one identity and one protocol surface.
- 7 deployment classes including air-gapped and edge-constrained.
- ML models and containers go through the same deployment gates.
Objections answered
- Haephestus is not a GitHub wrapper or a Harbor plugin — it is a ground-up Rust-native platform.
- Haephestus does not require cloud connectivity — air-gapped is a first-class deployment.
- Haephestus does not depend on upstream GitHub, GitLab, Harbor, or MLflow — it owns the entire surface.
Operator relevance: A DevOps lead can reduce security surface and audit scope by consolidating source control, registry, ML, and deployment into one platform with one identity.
Core Commitments
- RuntimeRust-native — no Go, no Python, no Node.js in the hot path
- Verticals4 — SCM/collaboration, Git engine/code search, container/artifact registry, ML tracking/model registry
- IdentityUnified across all verticals — OAuth 2.0, OIDC, SAML, SCIM, WebAuthn
- ProtocolOne REST API (OpenAPI 3.1), one gRPC (protobuf), one CLI, one SDK across all verticals
- RegistryOCI Distribution Spec v1.1.1 compatible — Docker, Kubernetes, Helm tools work without modification
- ML TrackingExperiment tracking, model registry, model versioning, evaluation, GenAI gateway
- Code SearchTrigram indexing across all repositories — find any pattern in seconds
- Deployment7 classes — self-hosted, BYOC, managed, shared SaaS, hybrid, air-gapped, edge-constrained
- AuditTamper-evident, cryptographically chained — one audit trail across all verticals
- SBOMSPDX and CycloneDX — supply-chain provenance for every artifact and deployment
- SDKsGo, Rust, TypeScript — first-party, typed, same methods across languages
- SecurityEvidence-bearing privileged operations, dual-approval classes, recovery-critical mutations
- Air-GappedFully disconnected with signed bundle updates and quantum-resistant signing
- UIOne web interface for all verticals — WCAG 2.2 compliant, deployment-class-aware
Platform
- RuntimeRust-native, single binary
- VerticalsSCM, registry, ML tracking, code search
- Deployment7 classes including air-gapped and edge
Integration
- IdentityUnified — OAuth, OIDC, SAML, SCIM, WebAuthn
- ProtocolOne REST, one gRPC, one CLI, one SDK
- CompatibilityOCI v1.1.1, Git, Helm, Docker — your tools work
Security and Compliance
- AuditOne tamper-evident trail across all verticals
- SBOMSPDX and CycloneDX supply-chain provenance
- Air-GappedSigned bundles, quantum-resistant signing