10-minute local proof

Prove your agent memory survives a restart.

Write one decision with a source, stop AllSource Prime, reopen the same data directory, then recall the decision and inspect the event that produced it. Local only. No account, API key, or telemetry from the Prime binary.

Run restart proof Read storage model

Pass 1

Recall returns the decision.

Pass 2

History returns its source event.

Pass 3

Both still pass after restart.

01Install Prime

Requires Rust 1.92 or newer. Prime runs as one local binary.

cargo install allsource-prime

02Start local HTTP mode

Run this in terminal A. Keep it open. Data stays under one explicit temporary directory so restart uses the same store.

RESTART_PROOF_DIR="${TMPDIR:-/tmp}/allsource-restart-proof"
mkdir -p "$RESTART_PROOF_DIR"
allsource-prime --mode http --port 3905 --data-dir "$RESTART_PROOF_DIR"

03Write, recall, inspect source history

Run this in terminal B. It creates one decision, adds searchable text, recalls it, and fetches its immutable event history.

RESTART_PROOF_NODE="$(
  curl -fsS -X POST http://127.0.0.1:3905/api/v1/prime/nodes \
    -H 'Content-Type: application/json' \
    -d '{"type":"decision","properties":{"question":"Which queue handles checkout processing?","answer":"durable-events","source":"runbook-42"}}' \
    | jq -r .entity_id
)"

curl -fsS -X POST http://127.0.0.1:3905/api/v1/prime/vectors \
  -H 'Content-Type: application/json' \
  -d "{"id":"$RESTART_PROOF_NODE","text":"Checkout processing moved to the durable-events queue because runbook-42 requires replay after restart."}"

curl -fsS -X POST http://127.0.0.1:3905/api/v1/prime/recall \
  -H 'Content-Type: application/json' \
  -d '{"text":"Which queue must survive a restart?","top_k":3}' | jq

curl -fsS "http://127.0.0.1:3905/api/v1/prime/nodes/$RESTART_PROOF_NODE/history" | jq

Before restart, recall should return a decision node. History should contain prime.node.created and the visible runbook-42source property.

04Restart, then run the same checks

Press Ctrl+C in terminal A. Run step 02 again without changing RESTART_PROOF_DIR. Then run these checks in terminal B.

curl -fsS -X POST http://127.0.0.1:3905/api/v1/prime/recall \
  -H 'Content-Type: application/json' \
  -d '{"text":"Which queue must survive a restart?","top_k":3}' | jq

curl -fsS "http://127.0.0.1:3905/api/v1/prime/nodes/$RESTART_PROOF_NODE/history" | jq

Proof passes when recall returns the same decision and history returns the same source-bearing creation event after restart. Score can vary by build and hardware; node identity and event history are the checks.

What this proves—and what it does not

Proves

Accepted local memory survives one clean process restart.
Semantic recall can find the stored decision after recovery.
History exposes the event payload carrying its source field.

Does not prove

Your production failure mode is fixed.
Crash durability beyond your chosen fsync policy.
Hosted availability, compliance, or workload-specific latency.

Now run your real failing memory flow.

Local proof establishes the storage behavior. Hosted AllSource lets you connect a real agent, keep tenant-scoped history, and test the same restart and provenance circuit during a 14-day trial.

Connect hosted memory Inspect source

No review, testimonial, endorsement, or public mention required.