Session Memory
Persist user goals, messages, edits, files, decisions, and long-running task state as time-indexed context records.
Flagship use case: Context
Context memory is the main reason TemporalStore exists.
TemporalStore should be read first as a context engine for AI agents: it stores session history, tool traces, retrieval evidence, summaries, safety counters, and replayable decision state. The other use cases are bonus examples of the same temporal store.
Context model
Agent memory needs temporal storage, not a loose prompt buffer.
Tool Traces
Record commands, tool calls, approvals, generated artifacts, and failures so an agent can resume with evidence.
Retrieval Evidence
Build compact ContextPacks from source-backed memories, summaries, timestamps, and relevance indexes.
Safety Counters
Keep rate limits, policy events, escalation history, and risky-action counters beside the memory stream.
Replayable Decisions
Explain what context was available when an agent answered, edited code, generated output, or asked for approval.
Iterative Workflows
Support open-ended sessions where a user keeps feeding prompts into the current thread and expects continuity.
Technical essay with Context plus supporting workloads
This is the detailed essay with the concrete use-case table and architecture narrative. It covers LLM agent context plus the bonus risk, fraud, ads, recommendation, and marketplace trust examples.
Read the detailed essayTemporal decision-state serving
This page explains the customer-facing product framing: online windows, filtered aggregates, distinct counts, behavior sequences, frequency caps, and AI context.
Read the TemporalStore product pageBonus examples
Other workloads show the same store handles real temporal state.
| Use case | Entity key | TemporalStore model | Online question |
|---|---|---|---|
| Purchase velocity | user_id |
TemporalCounter | How many purchases happened in the last 5 minutes? |
| Failed-login risk | device_id |
TemporalAggregate with dimensions | How many failed logins happened by country and method in the last 30 minutes? |
| Card testing | card_id |
TemporalDistinct | How many unique merchants did this card touch in the last 24 hours? |
| Chargeback monitoring | merchant_id |
TemporalAggregate | What chargeback counts and amounts are visible by channel in the last 7 days? |
| Ads frequency cap | campaign_id + user_id |
Composite-key counter | How many impressions happened in the last hour, day, or campaign window? |
| Recommendation sequence | user_id |
Sequence | What recent product, search, or content sequence should the ranker see? |
| Agent context | session_id |
Sequence plus counters | What tool calls, safety counters, and preference deltas should be available now? |
Data models
Context is the flagship model; other primitives support it.
TemporalCounter, TemporalAggregate, TemporalDistinct, Sequence, Profile, Context.
Benchmarks
Connect use cases to evidence and test artifacts.
Scale, parity, context quality, and benchmark source files.
Entry page
Return to the TemporalStore.AI technical homepage.
The homepage now links this page above the fold.