| --- |
| title: Orca Cognitive Runtime |
| emoji: 📈 |
| colorFrom: blue |
| colorTo: blue |
| sdk: static |
| pinned: false |
| license: apache-2.0 |
| short_description: 'ORCA: LLM agents should reuse cognition, not recompute it' |
| --- |
| |
| # LLM Agents Should Reuse Cognition, Not Reconstruct It |
|
|
| Most current LLM agent systems rely on recomputing tasks from scratch via prompts. |
|
|
| This works, but it does not scale. |
|
|
| It introduces clear limitations: |
| - lack of composability |
| - poor observability |
| - limited control |
|
|
| ## The problem |
|
|
| In any engineered system, complex tasks are not re-derived from first principles every time. |
|
|
| They are encoded into reusable procedures. |
|
|
| LLM agents today violate this principle. |
|
|
| ## The insight |
|
|
| Cognition should not be ephemeral. |
|
|
| It should be: |
| - structured |
| - reusable |
| - executable |
|
|
| ## ORCA: Cognitive Runtime for Agents |
|
|
| We introduce ORCA (Open Cognitive Runtime for Agents), a framework that separates cognition from execution. |
|
|
| It models cognition as: |
|
|
| - **Capabilities** → atomic cognitive operations |
| - **Skills** → structured workflows |
| - **Runtime** → execution layer |
|
|
| This enables: |
| - reuse of cognitive structures |
| - better control and observability |
| - composable agent systems |
|
|
| ## Links |
|
|
| - 📄 Paper: [arXiv link] |
| - 💻 Code: https://github.com/gfernandf/agent-skills# LLM Agents Should Reuse Cognition, Not Reconstruct It |
|
|
| Most current LLM agent systems rely on recomputing tasks from scratch via prompts. |
|
|
| This works, but it does not scale. |
|
|
| It introduces clear limitations: |
| - lack of composability |
| - poor observability |
| - limited control |
|
|
| ## The problem |
|
|
| In any engineered system, complex tasks are not re-derived from first principles every time. |
|
|
| They are encoded into reusable procedures. |
|
|
| LLM agents today violate this principle. |
|
|
| ## The insight |
|
|
| Cognition should not be ephemeral. |
|
|
| It should be: |
| - structured |
| - reusable |
| - executable |
|
|
| ## ORCA: Cognitive Runtime for Agents |
|
|
| We introduce ORCA (Open Cognitive Runtime for Agents), a framework that separates cognition from execution. |
|
|
| It models cognition as: |
|
|
| - **Capabilities** → atomic cognitive operations |
| - **Skills** → structured workflows |
| - **Runtime** → execution layer |
|
|
| This enables: |
| - reuse of cognitive structures |
| - better control and observability |
| - composable agent systems |
|
|
| ## Links |
|
|
| - 📄 Paper: https://zenodo.org/records/19438943 |
| - 💻 Code: https://github.com/gfernandf/agent-skills |
|
|
