Agent with persistent memory via remember()/recall()
Demonstrates: the remember/recall builtins backed by a [persistent] slot, a knowledge-base lifecycle, and recall() returning () for unknown keys (null-checked with ??). Run: SOMA_LLM_MOCK=echo soma run memory_agent.cell --signal learn "capital" "Paris" Test: SOMA_LLM_MOCK=echo soma test memory_agent.cell
A Soma program from the agents part of the verified corpus: 61 lines, cells Librarian, LibrarianTests. It passes soma check with soma 3.2.4.
How it is built
The cell Librarian declares 3 signals in its face: learn(key: String, value: String) -> Map, lookup(key: String) -> String and answer(question: String) -> Map; keeps 1 memory slot: kb (Map<String, String>, persistent). Its state machine task starts in idle and allows 3 transitions: idle -> working, working -> done and * -> failed. done and failed are terminal. Handlers: learn (line 27), lookup (line 33) and answer (line 38).
The file also holds 1 test cell (LibrarianTests) with 4 assertions, run by soma test.
The source
soma.toml
[package]
name = "agents-corpus"
version = "0.1.0"
description = "Verified AI-agent example cells (training corpus)"
# Agent LLM config. `mock = "echo"` keeps every example in this directory
# hermetic: `soma test` / `soma run` need no network and no API key, and
# results are deterministic. For a live run, delete the `mock` line (or
# override with SOMA_LLM_MOCK) and provide the key via env.
[agent]
provider = "anthropic"
model = "claude-opus-4-6"
key = "${ANTHROPIC_API_KEY}"
mock = "echo"
# Termination proof for verified_termination.cell's `workflow` machine:
# every path eventually reaches a terminal, no deadlocks.
# `cells` scopes these properties to Solver — the other agents in this
# directory have their own state names and must not inherit them.
[verify]
cells = ["Solver"]
deadlock_free = true
eventually = ["done", "failed"]
[verify.after.understanding]
eventually = ["solving", "failed"]
[verify.after.solving]
eventually = ["verifying", "failed"]
[verify.after.verifying]
eventually = ["done", "failed"]memory_agent.cell
// Agent with persistent memory via remember()/recall().
// Demonstrates: the remember/recall builtins backed by a [persistent] slot,
// a knowledge-base lifecycle, and recall() returning () for unknown keys
// (null-checked with ??).
//
// Run: SOMA_LLM_MOCK=echo soma run memory_agent.cell --signal learn "capital" "Paris"
// Test: SOMA_LLM_MOCK=echo soma test memory_agent.cell
cell agent Librarian {
face {
signal learn(key: String, value: String) -> Map
signal lookup(key: String) -> String
signal answer(question: String) -> Map
}
memory {
kb: Map<String, String> [persistent]
}
state task {
initial: idle
idle -> working
working -> done
* -> failed
}
on learn(key: String, value: String) {
remember(key, value)
kb.set(key, value)
map("stored", key, "value", value)
}
on lookup(key: String) {
let v = recall(key)
v ?? "unknown"
}
on answer(question: String) {
set_budget(2000)
let id = to_string(next_id())
transition(id, "working")
let r = try { think("Answer using stored facts: {question}") }
if r.error != () {
transition(id, "failed")
return map("status", "failed", "error", r.error)
}
transition(id, "done")
map("status", "done", "answer", r.value)
}
}
cell test LibrarianTests {
rules {
assert learn("capital", "Paris").stored == "capital"
assert lookup("capital") == "Paris"
// recall on an unknown key falls back via ??.
assert lookup("nope") == "unknown"
assert answer("hi").status == "done"
}
}What the toolchain says
$ soma check memory_agent.cell
✓ All checks passed.$ soma test memory_agent.cell
test LibrarianTests ...
✓ assert learn("capital", "Paris").stored == "capital"
✓ assert lookup("capital") == "Paris"
✓ assert lookup("nope") == "unknown"
✓ assert answer("hi").status == "done"
4 tests: 4 passed, 0 failed$ soma verify memory_agent.cell
Verifying memory_agent.cell...
note: soma.toml [verify] cells names 'Solver', which this file does not define — its properties are not checked here
State machine 'task': 4 states, initial 'idle'
States: [done, failed, idle, working]
✓ 4 states, 5 transitions
✓ all states reachable from 'idle'
✓ terminal states: [failed]
✓ no deadlocks
✓ liveness: every state can eventually reach a terminal state
✓ wildcard transitions: * -> [failed]
⚠ `* -> failed` also adds done -> failed — so done is not final. If it should stay final, write `* -> failed except [done]`
✓ think-isolated: CTL safety properties hold regardless of LLM output (1 handlers, 3 literal transitions, 0 dynamic)
✓ termination: all 3 handlers structurally terminate
✓ refinement: handler `answer` ⟶ {working, failed [if r.error != ()], done}
· transition sources: `answer` calls transition(id, target) without declaring the source — the runtime refuses a move the graph forbids, but the proof does not see where the handler starts; `transition(id, from, to)` names the edge (checked by `soma check`, refused at run time from any other state)
✓ effects: handler `answer` calls think() at 2 sites (the cost line counts loop rounds) — no tools available
10 passed, 1 warnings, 0 failures
Temporal properties for 'task':
✓ deadlock_free — no deadlocks in any reachable state
soma.toml: 4 user-defined properties loaded — NONE applies to this file ([verify] cells = ["Solver"] names no cell of it)
Temporal: 1 passed, 0 failed
VERIFY OKBuiltins used
map, next_id, recall, remember, set_budget, think, to_string, transition