Open-source platform for AI-assisted work

Give every task the context it needs.

Perseus adds current workspace context, durable memory, and reviewable records around the model or system you already use.

Local by default. You choose the model, data path, deployment, and permissions.

Product measurementsCompany-run A/B and scale tests
52.63%fewer full-request prompt tokensNaive full-file assembly vs. shipped-default render · cold cache · context + 14 prompts · named repo corpusRead the token report ↗
99.8%session recall@10LongMemEval-S current-main · retrieval only · 500 instancesRetrieval report ↗
40 writes/sdurable write throughputMeasured at 100K entities on named hardware · content-hashed scale report · not an SLARead the scale report ↗

Token reduction measures content assembly, not provider billing or dollar savings. Retrieval is retrieval-only; throughput is not an SLA.

Open evidence

See the result, then the method.

MemConflict tests changing facts, planted contradictions, and conditional preferences across eight self-hosted configurations.

MemConflict / tested Perseus Vault configuration0.555

Macro score in the self-run replication.

18 wrong across 3,750 questions; 1,378 were blank or abstaining (36.7%). Wrong and blank outcomes are counted separately.

Read the eight-provider report ↗

Deployment

Run it where the work is.

Local, on premises, private cloud, or disconnected. The program chooses the data, keys, network, and permissions.

See the security boundary →
Data
Stays in the environment you select for the local path.
Keys
You supply and keep the keys used for encryption and services.
Authority
The model and host keep their existing permissions. Perseus does not add mission-system authority.

Next step

See it work or start locally.