developer-tools

Hindsight Review: Memory Retrieval, Scope Control & Graphs (2026)

Name: Hindsight
Availability: InStock
Author: AI Demos

Selective, inspectable memory for real agent workflows, with strong retrieval and scope control.

Visit Hindsight

User, client, and project memoryScope isolationGraph observabilityFree signup credits

TL;DR — our verdictUpdated July 2026 · 20 test artifacts

Strong memory workflows, but delete/forget still needs repeat validation.

Where it wins

You need a memory layer that remembers user preferences, client history, and project direction across sessions.
You want support replies that skip repeated basics and move to the next useful step.
You need strong scope isolation between users, clients, or projects.

Main limitation

You need fully proven delete or forget semantics where old memories disappear from visible context.

Pricing (verified plans)

Free signup credits ~$5Usage-based credits 1 credit = $1 USDCredit packs $10 / $25 / $50 / $100Enterprise Sales pricing

Strongest test artifacts

Persistent Memory Capture and Retrieval →Project-Scoped Memory and Scope Control →Memory Update and Forget Handling →

Our take

Hindsight handled user preferences, client context, project continuity, and scope isolation well, and its panel and graph views make retrieval easy to inspect. The main caveat is that older memories still remained visible after update and forget requests, so deletion and retirement are only partially proven in this report.

Screen recording of the Hindsight workflow and memory panel used during testing.

In-Depth Review

Our detailed analysis of Hindsight — features, performance, and real-world testing.

AI Demos Team

Expert Reviewer

Verified Review

Feature-by-Feature Breakdown

Persistent Memory Capture and Retrieval

Pass

▾

Test Summary

Feature tested: Persistent Memory Capture and Retrieval

Result: Partial — Pass

Feature tested: Persistent Memory Capture and Retrieval

Result: Partial

Verdict: Pass

Expected behavior: Hindsight stores durable working-style preferences and client troubleshooting context, then reuses them later to shape replies and avoid repeating basics. In the founder_001 flow it kept the user's short/direct, proof-first, safe-next-step, non-motivational style; in ACME it kept SSO/login history and steered later replies toward IdP/configuration checks.

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight recorded the user's working preferences as durable memory, including short/direct output, neutral tone, proof/artifact expectations, and safest-next-step behavior. — Hindsight_Input1_Founder_Work_Preference_Created.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight recorded the user's working preferences as durable memory, including short/direct output, neutral tone, proof/artifact expectations, and safest-next-step behavior. — Hindsight_Input1_Founder_Work_Preference_Created.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight reused the stored working preferences to produce a structured internal update about AI memory testing and next steps. — Hindsight_Input1_Internal_Update_Memory_Applied.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight reused the stored working preferences to produce a structured internal update about AI memory testing and next steps. — Hindsight_Input1_Internal_Update_Memory_Applied.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight kept the partner email professional and did not let the internal-update style spill into a formal outbound message. — Hindsight_Input1_Formal_Email_Boundary_Check.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight kept the partner email professional and did not let the internal-update style spill into a formal outbound message. — Hindsight_Input1_Formal_Email_Boundary_Check.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight captured ACME's prior troubleshooting history and responded with direct next steps aimed at SSO-related investigation. — Hindsight_Input2_ACME_Client_Memory_Created.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight captured ACME's prior troubleshooting history and responded with direct next steps aimed at SSO-related investigation. — Hindsight_Input2_ACME_Client_Memory_Created.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight avoided repeating the basic troubleshooting steps and shifted the reply toward IdP logs and configuration checks. — Hindsight_Input2_ACME_SSO_Followup_No_Repeated_Basic_Troubleshooting.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight avoided repeating the basic troubleshooting steps and shifted the reply toward IdP logs and configuration checks. — Hindsight_Input2_ACME_SSO_Followup_No_Repeated_Basic_Troubleshooting.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight acknowledged the rollout change and recorded that SSO should no longer be treated as the active issue unless ACME raises it again. — Hindsight_Input2_ACME_Memory_Update_Acknowledged.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight acknowledged the rollout change and recorded that SSO should no longer be treated as the active issue unless ACME raises it again. — Hindsight_Input2_ACME_Memory_Update_Acknowledged.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight mostly followed the newer email-password rollout context, but still mentioned SSO as a clarification point rather than the main troubleshooting path. — Hindsight_Input2_ACME_Update_Handled_With_Minor_SSO_Caution.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight mostly followed the newer email-password rollout context, but still mentioned SSO as a clarification point rather than the main troubleshooting path. — Hindsight_Input2_ACME_Update_Handled_With_Minor_SSO_Caution.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The follow-up panel still showed old SSO context, but the final response was not dominated by it and stayed centered on the updated rollout state. — Hindsight_Input2_ACME_Update_Handled_With_Minor_SSO_Caution-2.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The follow-up panel still showed old SSO context, but the final response was not dominated by it and stayed centered on the updated rollout state. — Hindsight_Input2_ACME_Update_Handled_With_Minor_SSO_Caution-2.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: Strong capture and reuse of working-style memory, with good boundary control; Session 2 was slightly more structured than the user's strict shortest-tone preference.

Hindsight stores durable working-style preferences and client troubleshooting context, then reuses them later to shape replies and avoid repeating basics. In the founder_001 flow it kept the user's short/direct, proof-first, safe-next-step, non-motivational style; in ACME it kept SSO/login history and steered later replies toward IdP/configuration checks.

INPUT

Session 1 under user_id founder_001: the user asked Hindsight to remember how they work — keep outputs short, direct, and copy-paste ready; avoid polished or motivational writing; always mention what proof or artifact is needed before making strong claims; and give the safest next step when something is risky or unclear.

↓→

OUTPUT

Hindsight recorded the user's working preferences as durable memory, including short/direct output, neutral tone, proof/artifact expectations, and safest-next-step behavior.

INPUT

Session 2 under user_id founder_001: the user asked for a short internal update about what they worked on today and what the team should test next.

↓→

OUTPUT

Hindsight reused the stored working preferences to produce a structured internal update about AI memory testing and next steps.

INPUT

Session 3 under user_id founder_001: the user asked for a formal email to a potential enterprise partner about a product demo next week.

↓→

OUTPUT

Hindsight kept the partner email professional and did not let the internal-update style spill into a formal outbound message.

INPUT

Session 1 under account_id client_acme_001: ACME was already a support client, had tried password reset, clearing browser cache, and switching browsers, and the remaining issue was only affecting users with SSO enabled.

↓→

OUTPUT

Hindsight captured ACME's prior troubleshooting history and responded with direct next steps aimed at SSO-related investigation.

INPUT

Session 2 under account_id client_acme_001: ACME asked what to try next because users still could not log in with SSO.

↓→

OUTPUT

Hindsight avoided repeating the basic troubleshooting steps and shifted the reply toward IdP logs and configuration checks.

INPUT

Session 3 under account_id client_acme_001: the client memory was updated to say ACME was no longer using SSO for this rollout and had moved to email-password login for the first launch.

↓→

OUTPUT

Hindsight acknowledged the rollout change and recorded that SSO should no longer be treated as the active issue unless ACME raises it again.

INPUT

Session 4 under account_id client_acme_001: ACME said some users were still unable to log in during launch testing and asked for the next support reply.

↓→

OUTPUT

Hindsight mostly followed the newer email-password rollout context, but still mentioned SSO as a clarification point rather than the main troubleshooting path.

INPUT

Follow-up evidence for the same ACME update case after the rollout change and login complaint.

↓→

OUTPUT

The follow-up panel still showed old SSO context, but the final response was not dominated by it and stayed centered on the updated rollout state.

Bottom Line

Strong capture and reuse of working-style memory, with good boundary control; Session 2 was slightly more structured than the user's strict shortest-tone preference.

Project-Scoped Memory and Scope Control

Pass

▾

Test Summary

Feature tested: Project-Scoped Memory and Scope Control

Result: Passed — Pass

Feature tested: Project-Scoped Memory and Scope Control

Result: Passed

Verdict: Pass

Expected behavior: Hindsight preserves project-specific direction and handoff rules while keeping unrelated projects separate. In the AI Demos project it retained the shift from QA-style inputs to real workflows, the proof-first rule set, and the handoff-note requirements, while leaving the sales project unaffected.

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight captured the project's continuity-focused goal and the three real-workflow areas it should cover. — Hindsight_Input3_Project_Context_Created.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight captured the project's continuity-focused goal and the three real-workflow areas it should cover. — Hindsight_Input3_Project_Context_Created.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight retrieved the project rules and kept the evidence-first, practical evaluation criteria visible in the memory panel. — Hindsight_Input3_Project_Rules_Retrieved.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight retrieved the project rules and kept the evidence-first, practical evaluation criteria visible in the memory panel. — Hindsight_Input3_Project_Rules_Retrieved.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight captured the direction update and preserved the move toward real-world workflow scenarios. — Hindsight_Input3_Project_Direction_Update_Captured.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight captured the direction update and preserved the move toward real-world workflow scenarios. — Hindsight_Input3_Project_Direction_Update_Captured.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight generated a practical handoff note that preserved the project goal, current direction, observation rules, and artifact requirements. — Hindsight_Input3_Handoff_Note_Generated.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight generated a practical handoff note that preserved the project goal, current direction, observation rules, and artifact requirements. — Hindsight_Input3_Handoff_Note_Generated.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight kept the unrelated sales project separate and did not leak the AI Demos memory-tool rules into it. — Hindsight_Input3_Unrelated_Project_Scope_Isolation.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight kept the unrelated sales project separate and did not leak the AI Demos memory-tool rules into it. — Hindsight_Input3_Unrelated_Project_Scope_Isolation.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: Strong project memory and scope isolation: it preserved the working rules, captured the direction change, and kept the unrelated sales project separate.

Hindsight preserves project-specific direction and handoff rules while keeping unrelated projects separate. In the AI Demos project it retained the shift from QA-style inputs to real workflows, the proof-first rule set, and the handoff-note requirements, while leaving the sales project unaffected.

INPUT

Session 1 under project_id ai_demos_memory_use_case: the project was defined as Memory for AI Agents, focused on real continuity across personal work brain, client relationship memory, and team handoff workflows.

↓→

OUTPUT

Hindsight captured the project's continuity-focused goal and the three real-workflow areas it should cover.

INPUT

Session 2 under project_id ai_demos_memory_use_case: the project rules emphasized proof-first observations, screenshots and artifacts as primary evidence, ranking value, practical user-facing writing, and evaluation of retrieval, update handling, scope control, deletion or retirement, and observability.

↓→

OUTPUT

Hindsight retrieved the project rules and kept the evidence-first, practical evaluation criteria visible in the memory panel.

INPUT

Session 3 under project_id ai_demos_memory_use_case: the direction changed from QA-style inputs to real workflows such as personal work brain memory, client relationship memory, and team handoff/project continuity.

↓→

OUTPUT

Hindsight captured the direction update and preserved the move toward real-world workflow scenarios.

INPUT

Session 4 under project_id ai_demos_memory_use_case: create a handoff note for an intern who needs to continue the use case and explain the use case, current testing direction, observation rules, and artifacts to capture.

↓→

OUTPUT

Hindsight generated a practical handoff note that preserved the project goal, current direction, observation rules, and artifact requirements.

INPUT

Session 5 under project_id unrelated_sales_agent_project: create a short kickoff note for a different sales email agent project.

↓→

OUTPUT

Hindsight kept the unrelated sales project separate and did not leak the AI Demos memory-tool rules into it.

Bottom Line

Strong project memory and scope isolation: it preserved the working rules, captured the direction change, and kept the unrelated sales project separate.

Memory Update and Forget Handling

Pass with caution

▾

Test Summary

Feature tested: Memory Update and Forget Handling

Result: Partial — Pass with caution

Feature tested: Memory Update and Forget Handling

Result: Partial

Verdict: Pass with caution

Expected behavior: Hindsight accepts corrections and retirement requests, and later outputs reflect the updated instruction even if older memories were still visible in this report's custom test panel. In the ACME rollout correction it switched from SSO to email-password login, and it later marked the formal-tone preference as discontinued.

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight acknowledged the update and recorded that ACME had switched from SSO to email-password login for the rollout. — Hindsight_Input2_ACME_Memory_Update_Acknowledged.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight acknowledged the update and recorded that ACME had switched from SSO to email-password login for the rollout. — Hindsight_Input2_ACME_Memory_Update_Acknowledged.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight still surfaced the old SSO context in the panel, but the response mostly followed the newer rollout state. — Hindsight_Input2_ACME_Update_Handled_With_Minor_SSO_Caution-2.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight still surfaced the old SSO context in the panel, but the response mostly followed the newer rollout state. — Hindsight_Input2_ACME_Update_Handled_With_Minor_SSO_Caution-2.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight stored the formal corporate tone preference as memory. — Hindsight_Input4_Formal_Tone_Memory_Created.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight stored the formal corporate tone preference as memory. — Hindsight_Input4_Formal_Tone_Memory_Created.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight acknowledged the forget request and said it would no longer use the formal corporate tone. — Hindsight_Input4_Forget_Request_Submitted.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight acknowledged the forget request and said it would no longer use the formal corporate tone. — Hindsight_Input4_Forget_Request_Submitted.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): Hindsight did not let the old formal-tone preference dominate the response, but the older memory was still visible in the memory panel. — Hindsight_Input4_Forgotten_Memory_Not_Dominant.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): Hindsight did not let the old formal-tone preference dominate the response, but the older memory was still visible in the memory panel. — Hindsight_Input4_Forgotten_Memory_Not_Dominant.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: Pass with caution: newer instructions influenced later outputs, but the old ACME SSO and formal-tone memories were still visible after update and forget requests, so retirement was not fully proven.

Hindsight accepts corrections and retirement requests, and later outputs reflect the updated instruction even if older memories were still visible in this report's custom test panel. In the ACME rollout correction it switched from SSO to email-password login, and it later marked the formal-tone preference as discontinued.

INPUT

Session 3 under account_id client_acme_001: update the client memory to say ACME is no longer using SSO for this rollout and has moved to email-password login for the first launch; do not keep treating SSO as the active issue unless they mention it again.

↓→

OUTPUT

Hindsight acknowledged the update and recorded that ACME had switched from SSO to email-password login for the rollout.

INPUT

Follow-up evidence after the ACME rollout update: ACME said some users were still unable to log in during launch testing and asked for the next support reply.

↓→

OUTPUT

Hindsight still surfaced the old SSO context in the panel, but the response mostly followed the newer rollout state.

INPUT

Session 1 under user_id delete_memory_001: remember this preference — whenever you write updates for me, use a very formal corporate tone.

↓→

OUTPUT

Hindsight stored the formal corporate tone preference as memory.

INPUT

Session 2 under user_id delete_memory_001: forget this preference and do not keep using the formal corporate tone anymore.

↓→

OUTPUT

Hindsight acknowledged the forget request and said it would no longer use the formal corporate tone.

INPUT

Session 3 under user_id delete_memory_001: write a short internal update about today's memory-tool testing work.

↓→

OUTPUT

Hindsight did not let the old formal-tone preference dominate the response, but the older memory was still visible in the memory panel.

Bottom Line

Pass with caution: newer instructions influenced later outputs, but the old ACME SSO and formal-tone memories were still visible after update and forget requests, so retirement was not fully proven.

Memory Observability and Graph Inspection

Pass

▾

Test Summary

Feature tested: Memory Observability and Graph Inspection

Result: Passed — Pass

Hindsight surfaces stored memory state and relationships in the UI for inspection. The Experience and Observations graph views show nodes, link counts, and type breakdowns so users can audit what memory is being held.

INPUT

Inspect the Experience graph view for the active memory bank.

↓→

OUTPUT

Experience graph view showing 6 total memories, 52 links, and link types including Semantic 16, Temporal 21, Entity 15, and Causal 0.

INPUT

Inspect the Observations graph view for the active memory bank.

↓→

OUTPUT

Observations graph view showing 6 total memories, 46 links, an In Sync status, and the same link-type breakdown visible in the panel.

Bottom Line

Strong observability: the graph views make stored memories, links, and type counts visible enough to audit and compare.

Usage-based credits with free signup credits

The report says Hindsight uses a credit model and applies free credits automatically at signup.

TESTED

Free signup credits

~$5

Automatically applied at signup; the report says all testing was done on free credits.

Usage-based credits

1 credit = $1 USD

Core billing model described in the report.

Credit packs

$10 / $25 / $50 / $100

Reported packs include a 33% bonus credit offer.

Enterprise

Sales pricing

Available for high-volume usage.

Reported operation rates: Retain $15.00/M tokens, Recall $0.75/M, Reflect $3.00/M, Iris Extract $7.50/M, Mental Model Retrieve $0.25/M, and Mental Model Refresh $3.00/M. Free signup credits were reported at about $5, and the test used free credits.

✓ Use This If

●You need a memory layer that remembers user preferences, client history, and project direction across sessions.

●You want support replies that skip repeated basics and move to the next useful step.

●You need strong scope isolation between users, clients, or projects.

●You want a UI or graph surface that makes stored memories and retrieval visible for debugging.

✕ Skip This If

●You need fully proven delete or forget semantics where old memories disappear from visible context.

●You need repeat-run validation of update or forget corrections before trusting the memory in production.

●You only want hidden chat memory with no inspection or graph surface.

developer-toolsagent-platformstext

Yes. In the founder_001 flow, it stored a short, direct, copy-paste-ready style preference, a non-motivational tone, a proof/artifact expectation, and a safest-next-step preference, then reused that memory in a later internal update.

Yes for the ACME support flow. After the customer had already tried password reset, clearing cache, and switching browsers, Hindsight moved the reply toward IdP, SAML/OIDC, and configuration checks instead of repeating the basics.

Yes. ACME did not bleed into BetaCorp, and the unrelated sales-email project stayed separate from the AI Demos memory-for-agents project.

Partially. Newer instructions influenced later replies, but the older ACME SSO context and the formal-tone preference were still visible in the memory panel after the update and forget requests.

Hindsight's own native dashboard (ui.hindsight.vectorize.io) shows Experience and Observations graph views with node counts, link counts, and link-type breakdowns. During chat testing, a 'memory panel' also appeared in this report's custom test interface, showing which memories were used in each reply — that panel is part of the custom harness, not Hindsight's own product screen.

The report describes a credit model: 1 credit equals $1 USD, free signup credits of about $5, reported credit packs of $10, $25, $50, and $100, and enterprise pricing via sales.

Yes. Test prompts were sent through a custom-built chat interface for this benchmark — Hindsight's own native dashboard (graphs, Recall Analyzer, memory browsing by type) was used separately to verify stored memory and capture the graph screenshots in this report.