"Intelligence is not generation. It is structured accumulation."
Part of the [&] Protocol stack · Ecosystem overview · Three-protocol stack · Stack status
Welcome to the documentation hub for OpenSentience — the research arm and runtime governance layer of the [&] Protocol ecosystem.
OpenSentience is not a product. It is three things:
A research organization that publishes cognitive protocols — formal
specifications grounding AI agent architecture in cognitive science.
A thin governance shim (open_sentience hex package) that enforces
permission, lifecycle, and autonomy policies on OTP-supervised agents.
The home of two cross-cutting protocols — OS-009 PRISM (diagnostic
benchmark) and OS-010 PULSE (loop manifest standard) — that sit above the eight cognitive primitives and turn the [&] portfolio into a measurable, composable, accountable agent substrate.
The research output is ten numbered protocols organized in two layers:
Cognitive primitives (OS-001 through OS-008) — one capability each
Cross-cutting protocols (OS-009 PRISM, OS-010 PULSE) — diagnostic + temporal layers above the primitives
The runtime output includes the Elixir governance package implementing OS-006 (shim) and OS-008 (harness), the PRISM benchmark engine (/PRISM/, Elixir/OTP, Fly.io, 6 MCP machines), and the PULSE manifest standard (/PULSE/, JSON Schema + reference manifests).
| Protocol | Name | Cognitive Basis | Implemented By |
|---|---|---|---|
| OS-001 | Continual Learning | Hippocampal consolidation | Graphonomous |
| OS-002 | Topological Routing (κ) | Prefrontal gating | Graphonomous routing layer |
| OS-003 | Deliberation Orchestrator | Dual-process theory | AgenTroMatic |
| OS-004 | Attention Engine | Endogenous attention | Graphonomous attention module |
| OS-005 | Model Tier Adaptation | Resource rationality | Graphonomous / Agentelic |
| OS-006 | Agent Governance Shim | Executive function | open_sentience hex package |
| OS-007 | Adversarial Robustness | Immune system | OpenSentience security module |
| OS-008 | Agent Harness | Supervisory attentional system | OpenSentience harness module |
| Protocol | Name | Layer | Implemented By |
|---|---|---|---|
| OS-009 | PRISM — Protocol for Rating Iterative System Memory | Diagnostic algebra (measures loops over time) | /PRISM/ Elixir/OTP, Fly.io, 6 MCP machines, prism.opensentience.org |
| OS-010 | PULSE — Protocol for Uniform Loop State Exchange | Temporal algebra (declares how loops cycle) | /PULSE/ manifest standard, JSON Schema, reference manifests, pulse.opensentience.org |
Each cognitive primitive maps a well-established cognitive science finding to a concrete software architecture. The cross-cutting protocols sit above them: PRISM measures how well a closed memory loop performs over time; PULSE declares how any loop in the [&] ecosystem cycles, nests, and signals across boundaries. Both are independent of one another and independent of [&] — a system may adopt one without the others.
Homepages
Root Docs
OpenSentience Docs
Skills
If you are new to OpenSentience, follow this path:
This page — understand scope and structure
positioning — where OpenSentience fits in the agent governance landscape
architecture — how the governance shim is built (OTP internals)
quickstart — install the hex package and run your first governed agent
ecosystem-overview — how all ten protocols connect across the [&] ecosystem
harness-engineering — the emerging discipline of harness design and OS-008's position
runtime-walkthrough — follow a concrete agent through governance + harness orchestration
competitive-landscape — how OpenSentience compares to alternative approaches
comparison-table — quick-reference feature comparison
research — cognitive science grounding for each protocol
faq — fill in remaining questions
skills/ — hands-on implementation skills for the governance shim
For deep protocol specifics:
spec/README — the full technical specification (all ten protocols)
spec/OS-008-HARNESS — the Agent Harness Protocol specification
spec/OS-009-PRISM-SPECIFICATION — PRISM diagnostic benchmark protocol
spec/OS-010-PULSE-SPECIFICATION — PULSE loop manifest standard
spec/OS-E001-EMPIRICAL-EVALUATION — empirical evaluation of topology-aware continual learning
Linked index of every page in this set (renders on GitHub and in the docs atlas; the toctrees above drive the Sphinx build):
Guides
Positioning — The Missing Constraint: Runtime Governance in the Agent Stack
Competitive Landscape — Agent Governance and Runtime Enforcement
Specifications
UX & build prompts
OpenSentience provides three layers of runtime enforcement:
OS-008 Harness: "Has this agent followed the correct process?"
Pipeline ordering, quality gates, sprint contracts
OS-006 Shim: "Is this agent allowed to do this?"
Permissions, lifecycle, autonomy levels
OS-007 Security: "Is this agent under attack?"
Threat detection, identity verification, circuit breakersAll three log to the same append-only audit trail, creating a unified compliance record from protocol to permission to pipeline.
Above the cognitive primitives, OpenSentience publishes three sibling protocols that together turn the [&] portfolio into a measurable, composable, accountable agent substrate:
┌──────────────────────────────────────────────────────────┐
│ PRISM (OS-009) — measures loops over time diagnostic │
├──────────────────────────────────────────────────────────┤
│ PULSE (OS-010) — declares loops + signals temporal │
├──────────────────────────────────────────────────────────┤
│ [&] Protocol — composes capabilities structural│
└──────────────────────────────────────────────────────────┘| Layer | Question | Artifact |
|---|---|---|
| [&] | What can each agent do, and how do they compose? | *.ampersand.json |
| PULSE | How do their processes cycle and signal each other? | *.pulse.json (loop manifest) |
| PRISM | How well do those cycles actually work over time? | Diagnostic reports + leaderboards |
A loop is PULSE-conforming if its manifest validates against pulse-loop-manifest.v0.1.json and its runtime passes the 12-test conformance suite. A system is PRISM-evaluable automatically once it is PULSE-conforming — PRISM's compose phase reads the manifest, injects scenarios at the declared retrieve boundary, and observes outcomes via the declared learn phase. Adoption order is typically [&] → PULSE → PRISM.
OpenSentience occupies the governance and research layer of the [&] stack:
[&] Protocol — capability composition (what an agent CAN do)
OpenSentience — governance + harness (what an agent MAY do, and how)
Graphonomous — memory + learning (OS-001, OS-002, OS-004, OS-005)
AgenTroMatic — deliberation (OS-003)
Delegatic — policy authoring (feeds into OS-006 enforcement)
Agentelic — agent manifests (consumed by OS-006 install)
FleetPrompt — marketplace (agents enter governance on install)
SpecPrompt — acceptance criteria (consumed by OS-008 quality gates)The governance shim wraps any OTP child_spec. It does not replace the supervised process — it interposes permission checks, lifecycle management, autonomy control, pipeline enforcement, and audit logging around it.
OpenSentience exists because intelligence without governance is generation without accumulation. The ten protocols formalize what structured accumulation requires: memory that consolidates, routing that detects cycles, deliberation that resolves conflict, attention that prioritizes, resources that adapt, governance that enforces boundaries, security that detects threats, a harness that orchestrates the entire pipeline, diagnostics that measure whether the loop actually learns (PRISM), and a temporal manifest that lets every loop in the ecosystem declare how it cycles and signals (PULSE).
The shim is deliberately thin. Governance should cost less than 1% of the system it governs. PULSE is a manifest standard with no required runtime. PRISM runs as a separate benchmark engine that any system can opt into via its PULSE manifest.