Research Preview — APIs may change. GitHub
Architecture
Astral runs geocomputation inside a self-contained Docker container, designed to execute within a Trusted Execution Environment (TEE) via EigenCompute. This page describes how the system is built.Execution model
Container design
Self-contained container
Self-contained container
PostGIS runs inside the Docker container, not as an external service. This is essential for verifiable computation in the TEE — no external dependencies means the entire execution environment is attested.
Stateless model
Stateless model
Each request brings all required inputs. No persistent state between requests. This ensures determinism and simplifies verification — same inputs always produce same outputs.
Signing key inside TEE
Signing key inside TEE
The service holds a signing key that is generated within the TEE or securely provisioned. The design intent is that the operator cannot extract it — a property that holds when the enclave runs under remote attestation (see deployment status above). All signed results are produced with this key.
Internal computation flow
PostGIS uses GEOS for geometry operations — the same C++ library used by QGIS, GDAL, and most professional geospatial software.
Why this architecture
The design choices above serve a single goal: making geocomputation results verifiable.- Self-contained means the TEE attestation covers the entire execution environment. No external database calls that could be intercepted or altered.
- Stateless means determinism is straightforward. Given the same inputs, the container produces the same output every time.
- Key inside TEE means that, when the enclave runs under attestation, the signing key cannot be extracted or used outside it. If you trust the TEE — and the enclave is attested — you trust the signature. Today, a valid signature proves a key Astral controls produced the result; binding that key to a continuously attested enclave is target-state, not a live guarantee.
Next: What is verified
What the signature covers and what computation reproducibility means