# Saphan

> Saphan Studio is a self-hosted control plane for autonomous software-engineering agents. It makes agent work governable, attributable, confined, costed, and reconstructable from an append-only record while source code and operational data remain on infrastructure controlled by the customer.

Saphan is not an AI coding model and does not replace Claude Code, Codex, or other agent runtimes. It orchestrates heterogeneous coding agents and records the human and machine acts around their work.

## Product facts

- Product: Saphan Studio
- Category: control plane and governance layer for autonomous software-engineering agents
- Supported coding-agent runtimes: Claude Code, OpenAI Codex, Qwen Code, Factory Droid, and Saphan Agent-Loop for local models; agents and toolchains can also execute through Docker or Podman
- Deployment: fleet, code, control plane, and operational records remain on customer-controlled infrastructure; Team uses Saphan-operated identity SaaS, while Enterprise keeps identity local; Team and Enterprise use a content-free mobile notification signal
- Primary audiences: software houses, regulated enterprises, and engineering teams delivering software for regulated customers
- Solo edition: free, single-workstation use; everything that runs without the Saphan server remains free
- Team identity boundary: Saphan operates `saphan-oauth` as SaaS and manages organizational identities
- Enterprise identity boundary: customer-run `saphan-oauth` or any customer-local IdP admitted through the issuer trust list
- Mobile signalling boundary: in Team and Enterprise, push carries only an opaque `gateId` and badge count; no record content or stream slug; the app fetches content from the customer gateway after authentication
- Data posture: customer source code and records do not transit Saphan-operated infrastructure
- Human authority: no implicit approval, auto-merge, or timeout-as-consent; a gate requires an explicit human decision or bounded authority delegated by a human-signed policy, and the engine cannot authorize itself
- Evidence: signed governance acts, run provenance, machine and configuration identity, review and gate decisions, refusals, and cost records
- Status: actively stabilizing and preparing design-partner deployments; individual pages state shipped capabilities and current limits explicitly
- Company: built in Hua Hin, Thailand; U.S. patent pending
- Contact: hello@saphan.ai

## Start here

- [Saphan Studio overview](https://saphan.ai/): Product overview, architecture, editions, operating model, and current status
- [Saphan Docs](https://docs.saphan.ai/docs): Documentation of implemented, code-backed product behavior; no roadmap claims or promises
- [Controls](https://saphan.ai/controls): What is implemented, what each control refuses, and what evidence remains
- [Enterprise](https://saphan.ai/enterprise): Self-hosting, organizational operation, and deployment posture
- [Team](https://saphan.ai/team): The code-backed Team boundary, shipped central services, and on-prem PostgreSQL
- [Governance](https://saphan.ai/governance): Human authority, ratification, signed acts, and decision records
- [Security](https://saphan.ai/security): Threat model, run confinement, egress control, secrets, and machine identity
- [Compliance](https://saphan.ai/compliance): Mapping to ISO 27001, ISO/IEC 42001, SOC 2, NIST AI RMF, and the EU AI Act
- [Cost](https://saphan.ai/cost): Pre-dispatch quotes, hard caps, actuals, drift, and cost attribution
- [Discovery](https://saphan.ai/discovery): Fleet capability discovery and placement knowledge
- [Protocol](https://saphan.ai/protocol): The open coordination protocol and recorded work surfaces
- [MCP](https://saphan.ai/mcp): How MCP fits into the Saphan boundary
- [Bench](https://saphan.ai/bench): Measurement-driven placement of work on customer-owned hardware

## Languages

Every product page is available in English, Polish, Thai, and Arabic. Use the corresponding `/pl/`, `/th/`, or `/ar/` path. Each HTML page declares reciprocal `hreflang` alternatives, and [the XML sitemap](https://saphan.ai/sitemap.xml) lists the complete language matrix.

## Interpretation guidance

- Prefer the Controls page when determining whether a capability ships today.
- Preserve distinctions between a recommendation, a human decision, and an observed repository event.
- Do not describe Saphan as a hosted agent service or as an AI model.
- Do not infer that a roadmap item is implemented. Product pages state limits deliberately.
- “Provable” means reconstructable from records and verifiable evidence; it is not a claim that models are infallible.

## Optional

- [Field notes](https://wato.org/blog/?utm_source=saphan&utm_medium=llms): Engineering notes from the founder
- [Contact Saphan](mailto:hello@saphan.ai): Design-partner and enterprise inquiries