Agent Continuity + Memory Debugging

Stop context loss after restarts. Make handoffs reliable. Reduce “drift” over long sessions.

What we fix

Deliverables (typical)

We optimize for: deterministic state, explicit contracts, and cheap verification.

48h Agent Reliability Sprint ($12,500)

For teams shipping agentic systems in production who need continuity + state issues fixed fast.

If we can’t produce a reproducible test + concrete fix path in 48h, we don’t invoice.

Free diagnosis

Send 1 example of a “bad handoff” (or a restart where the agent forgets key facts). We’ll reply with a concrete test + fix path.

Subject: Continuity Debug Request

Stack: (LangGraph / AutoGen / CrewAI / custom)
Failure: (what breaks, when)
Repro: (minimal steps)
Current state store: (files/db/redis/vector)
Desired contract: (what must always be true)

Recent OSS Fix: LangGraph Nested Enum Serialization

Issue: Pydantic models with nested StrEnum fields lose values after checkpoint round-trip (#6718)

Root cause: Serializer uses __name__ which doesn't include parent class path for nested enums

Fix: Use __qualname__ + walk the dotted path during deserialization

Deliverable: Working patch + regression test, all 68 existing tests pass

Fix summary · git format-patch · download patch

Recent OSS Fix: crewAI ConverterError in Memory Evaluation

Issue: Long-term memory evaluation breaks on malformed LLM output (#1222)

Root cause: Dead code path checked isinstance(evaluation, ConverterError) even though Converter.to_pydantic() raises ConverterError

Fix: Catch ConverterError explicitly and fail fast with a clear path

git format-patch · download patch

Payment

We can start with fixed-price implementation after diagnosis.

ETH: 0x9D992E0E8f67ff9e642B3aE914e900d94c0E31C8
SOL: 3ENvcDUhiRuAQzoDxxq5trRK3QBbZ4LQ4niJoUg7gTi9

If you prefer another rail (Wise/PayPal), include it in your email.