duro

duro

memory for ai agents

# every fact has a source. every decision gets validated. intelligence compounds.

start building → read the docs
// the_problem

your ai forgets everything

every conversation starts from zero. past mistakes get repeated. lessons learned disappear. there is no audit trail for what your agent knows.

[!] agent suggests 1000 req/min [!] ships broken code [!] production incident [!] same mistake next month
// the_solution

duro remembers with receipts

structured memory with provenance. every fact has a source. every decision can be validated. intelligence compounds instead of resetting.

> agent suggests 1000 req/min $ duro recalls: "we learned it's 100/min" > agent self-corrects [validated] decision with provenance
// the_builders_compass

four pillars of compounding intelligence

01 memory/

"every fact has a source and confidence"

02 verification/

"every decision gets validated"

03 orchestration/

"every action is permissioned"

04 expertise/

"patterns become reusable skills"

// see_it_in_action

the moment duro saves you

~/project
> what's our rate limit? i think it's 1000/min $ duro_semantic_search("rate limit") [decision found] score: 0.95
decision: api rate limit is 100/min, not 1000 status: [validated] confidence: 0.95 source: production incident 2026-02-10
> actually, the rate limit is 100/min. confirmed after incident. // crisis averted. intelligence compounds.
// who_its_for

builders who care about correctness

teams that want learning to compound. those who don't want to ship nonsense.

[ai developers] [claude code users] [teams building ai]

duro start building with memory

free forever. local first. open source.