Use cases
Five places where context goes missing.
Each case carries its status before its argument: one is demonstrated, the other four are working hypotheses. No quantified result is claimed — nothing has been measured in production yet.
01 · Demonstrated
Agent handoff
Status: demonstrated
19 checks out of 19, including a handoff to an agent declared on a different model. Replayable: `python demos/agent_handoff_demo.py`.
Problem
Agent A understood the case. It ends. Agent B restarts with a prompt and, at best, a transcript to re-read.
With Venkai
A wrote its decisions and constraints as it went. B calls recall() and receives what matters, ranked, without re-reading history.
Why it matters
It is the only one of the five cases that rests on no projection: the handover was executed, not assumed.
02 · Hypothesis
Long-running workflows
Status: hypothesis
The pieces are tested (state outside the session, versioning). No multi-day workflow has been measured end to end at a customer yet.
Problem
A workflow spread over several days runs into interruptions, resumptions and saturated context windows.
With Venkai
State lives outside the session. Resuming means asking for progress and next actions, not rebuilding history.
Why it matters
A larger context window pushes the limit back; it doesn't survive the end of the session.
03 · Hypothesis
Multi-agent software engineering
Status: hypothesis
Storing a decision and its reason is tested. That it actually stops one agent from undoing another's work remains to be shown on a real repository.
Problem
One agent picks a library, another replaces it three steps later without knowing why the first was chosen.
With Venkai
The decision and its reason are written as a typed object. The next agent can ask for architecture decisions before making one.
Why it matters
The cost of lost context isn't the token spent, it's the work undone.
04 · Hypothesis
Human → agent continuity
Status: hypothesis
Technically, nothing distinguishes a constraint written by a human from one written by an agent. The usage itself has not been observed in a team.
Problem
A human frames the work in a meeting or a review. The executing agent only sees what someone remembered to paste into its prompt.
With Venkai
Constraints and preferences set by the human are written once and read by every agent that follows.
Why it matters
The human constraint is the most expensive one to lose, because nobody restates it.
05 · Hypothesis
RAG plus operational context
Status: hypothesis
The separation is an architectural choice, not a measured result. No quantified comparison against RAG alone has been produced.
Problem
RAG retrieves what documents say. It doesn't know what the system already decided or what is left to do.
With Venkai
RAG stays the knowledge source; Venkai keeps the state of the work done with that knowledge. The two are read separately.
Why it matters
Writing work state into the document index mixes two kinds of information and degrades both.
Build a continuity layer for your agents.
Describe your multi-agent workflow and where context goes missing. We answer with a pilot scope, or with the reason Venkai isn't the right piece.
Describe your workflow