Instinct
Spear Street Technology, Inc. d/b/a Instinct
As of 9 Oct 2026, PAEval has compiled 15 capability records for Instinct from 5 source URLs; 3 are independent observations and 7 come from the vendor's own documentation. Assistant Benchmark (updated 8 Oct 2026) gives it 8.5 on its 1–10 scale, the mean of the 12 of 15 dimensions it scored, ranking #2 of 12 scored products. In micro1 PersonalAgentBench it completed 4 of 10 workflows without a critical violation (40% trusted completion; 95% interval 17–69%), and 4 of 10 counting any completion. Scores from different evaluators use different scales and are not combined. PAEval does not publish its own score or rank for it.
Official positioning and third-party action records are consistent with personal agency; it is not just a Q&A tool
- Independent3
- Vendor7
- Third-party report2
- Not established3
15 records · 5 source URLs · 1 external result(s)
Capability summary
Strongest status per domain; every underlying record is listed below. Availability does not establish task success.
Positioning & scope
Use text messages/phone calls to handle daily affairs; the official website provides examples of bill disputes, gifts, groceries, appointments and travel.
Regions, eligibility & access
Listing site records are invitation/waiting list, free during private testing; the official website public page does not provide a standard price list.
Conditions: Whether the invitation is available depends on the status after logging in.
Limitations: Free private testing cannot be considered free forever
The listed sources were reviewed, but sufficient public documentation was not found.
Limitations: We have checked the listed official website, help/FAQ or policies and third-party records, and found no sufficient public explanation; unknown does not mean that it is not supported.
Autonomy & background work
During the observation in September, duplicate car rentals and abnormal expenses were proactively discovered; the 7 o'clock briefing on the working day of September 11 was on time for two consecutive mornings.
Limitations: The briefing is test; the active discovery is observed, which is not the same type of standardized evidence.
The listed sources were reviewed, but sufficient public documentation was not found.
Limitations: We have checked the listed official website, help/FAQ or policies and third-party records, and found no sufficient public explanation; unknown does not mean that it is not supported.
Execution environments
The official website states that phones and computers can be used; the privacy policy covers Mac OS and mobile applications.
Limitations: No disclosure of cloud computer OS, device persistence, user-native operating scope, or self-hosting capabilities
Calls, messages & identity
Actions can be performed based on user authorization to connect accounts and credentials; clause prompt actions can bring financial/contractual consequences.
Limitations: The allocation method of independent mailboxes/numbers and whether pure read-only permissions can be selected are not fully disclosed by the official website.
Connectors & tools
Policy columns include Gmail, Calendar, Drive, Docs, Sheets, Slides, Tasks and related content, which can be used for agency processing.
Memory & personalization
The policy says to provide personalized recommendations based on interactions and historical experience.
Channels & interfaces
Officially supports text/call; iMessage, WhatsApp, and Phone are supported.
Research & deliverables
The listed sources were reviewed, but sufficient public documentation was not found.
Limitations: The policy mentions audio/voice and Google document access, but this does not prove that it can generate and edit all office formats; the official website checked does not have complete capabilities to help.
Collaboration & parallelism
The ranking site Multiplayer/groups has a score of 9 but no test/observed tag; the official details are not found on the public official website.
Reliability & recovery
Travel, shopping, and email on the list have action results; chain tasks are not scored, and the reliability of complex tasks cannot be derived from a single performance.
Permissions, security & data
Non-Google data can be used for model improvement/training with backward exit and security clearance exceptions; Google Workspace data is not used for training. Disconnecting does not automatically delete indexed data.
The terms and conditions may include confirmation/protection, but there is no guarantee that incorrect actions will be prevented; action records may also be inaccurate.
External measurements & hands-on results
Different methods, samples and dates. Not combined into one success rate.
Assistant Benchmark ↗
- Tested
- 2026-09-11
- Version
- v0.2
- Source score
- 8.5
- Scored dimensions
- 12
- Evidence labels
- test: 1; observed: 10; unlabelled: 1