Making conversational posture explicit and inspectable.
Purpose Intelligence (Pi) explores an external control layer around existing language models. v1.5 separates working estimation, routing, session state, interpretation, and presentation.
П Purpose Intelligence v1.0
A prompt-based interface developed while the Purpose question was still taking shape.
П Purpose Intelligence v1.5
An overlay with working estimates, routing, session state, guards, and guidance modes.
П Purpose Intelligence v2.0
A private research direction exploring more stateful, longer-horizon intent governance.
Inspectable conversational control
Pi does not expose a model's internal reasoning. It makes parts of the authored control layer easier to inspect: working state, routing, guidance posture, guards, and presentation.
Pi — Purpose Intelligence
Pi v1.5 is the current reference prototype for Purpose's conversational research. It combines working estimates, routing, session state, guards, interpretation, and presentation around an existing language model so those responsibilities can be examined separately.
Pi v1.5's calculations and guidance rules are inspectable as authored mechanisms. Whether the working estimates measure the properties their names suggest, and whether the overlay produces a meaningful effect relative to the same model without it, remain open empirical questions.
The Evolution of Pi
Pi is not a foundation model. It's a conversational research prototype built around existing language models to study how posture, routing, state, and continuation can remain explicit.
Pi v1.5 was developed around existing language models to separate responsibilities that can otherwise disappear into the final response: working estimation, routing, session state, guards, interpretation, and presentation.
As foundation models became more capable, the Purpose question became more specific. The research is not whether Pi can make a model generally more capable. It is whether making these responsibilities explicit outside the model changes how uncertainty is handled in ways that can be inspected and tested.
Pi v1.5 provides a working implementation for asking that question. Its authored source makes the control structure inspectable, while the retained probe records show that similar final handling can still emerge through different calculations and execution paths.
The next evidence requires a matched comparison: the same underlying model with and without a defined Purpose overlay, under the same inputs and outcome criteria.
→ Probes | System | R&D Log
As foundation models became more capable, the Purpose question became more specific. The research is not whether Pi can make a model generally more capable. It is whether making these responsibilities explicit outside the model changes how uncertainty is handled in ways that can be inspected and tested.
Pi v1.5 provides a working implementation for asking that question. Its authored source makes the control structure inspectable, while the retained probe records show that similar final handling can still emerge through different calculations and execution paths.
The next evidence requires a matched comparison: the same underlying model with and without a defined Purpose overlay, under the same inputs and outcome criteria.
→ Probes | System | R&D Log
The private v2 research direction explores what changes when governing context must persist beyond a single conversational moment: longer-horizon state, authority, continuation, and the conditions under which later action may inherit what came before.
This remains prospective research. A stateful implementation could make longer-horizon continuity available for study, but persistence itself would not establish coherence, correct interpretation, or appropriate authority.
A later governance scaffold therefore extends the Purpose question rather than completing it. These capabilities remain research directions, not established v1.5 behavior.
→ Probes | System | v2 Scaffold
This remains prospective research. A stateful implementation could make longer-horizon continuity available for study, but persistence itself would not establish coherence, correct interpretation, or appropriate authority.
A later governance scaffold therefore extends the Purpose question rather than completing it. These capabilities remain research directions, not established v1.5 behavior.
→ Probes | System | v2 Scaffold
Early prototype responses
The following are unsolicited individual reactions to the early v1 prototype. They are not a user study, a representative sample, or evidence that Purpose produces a particular outcome.
