Oris is an AI Self-Evolution Framework for closed-loop software improvement.
Most systems can execute tasks, but cannot systematically improve how they solve recurring problems.
Oris focuses on closed-loop improvement for AI software:
- Capture real signals from failures and regressions.
- Generate and validate candidate mutations safely.
- Promote proven solutions into reusable assets.
- Reuse successful solutions with confidence-aware replay.
| In Scope (Primary) | Supporting Layer | Not Primary in This README |
|---|---|---|
| Self-evolution loop and framework primitives | Durable execution and orchestration runtime | Exhaustive runtime API reference |
| Gene/Capsule lifecycle and confidence reuse | Storage/checkpoint backends and deployment integration | Detailed access policy, endpoint, and metrics contracts |
| Evolution-oriented scenario workflows | Production operations and integration surface | General-purpose workflow engine positioning |
Oris implements an 8-stage self-evolution loop:
- Detect — collect actionable signals from compiler/test/runtime outcomes.
- Select — choose the best candidate gene or strategy.
- Mutate — generate candidate changes from prior successful patterns.
- Execute — run mutations in a controlled sandbox.
- Validate — verify correctness and safety gates.
- Evaluate — compare improvement versus regression.
- Solidify — promote successful mutations into durable assets.
- Reuse — replay proven assets with confidence tracking.
Oris maps EvoMap concepts to concrete framework behavior:
| EvoMap Concept | Oris Mapping |
|---|---|
| Worker Pool | EvolutionPipeline stages |
| Task Queue | Signal intake and selection flow |
| Bounty System | Issue intake and prioritization |
| A2A Protocol | oris-evolution-network experimental protocol |
- Self-improving AI agents that learn from failed runs.
- Supervised or autonomous dev loops for recurring issues.
- Evolution-aware replay pipelines with confidence lifecycle.
- Cross-agent knowledge exchange over an evolution network surface.
Install the core crate and enable the framework surface:
cargo add oris-runtime
cargo add oris-runtime --features full-evolution-experimental
export OPENAI_API_KEY="your-key"Run the canonical evolution scenario:
cargo run -p evo_oris_repoRun the first-run script with observable artifacts:
bash scripts/evo_first_run.shExpected outputs:
target/evo_first_run/summary.jsontarget/evo_first_run/run.log
Maturity below reflects the current checked-in framework surface.
| Component | Crate | Maturity | Gate |
|---|---|---|---|
| Evolution Core | crates/oris-evolution |
Implemented baseline with extended lifecycle primitives | evolution-experimental |
| Sandbox | crates/oris-sandbox |
Implemented baseline | evolution-experimental |
| EvoKernel | crates/oris-evokernel |
Implemented baseline with governor-aware capture | evolution-experimental |
| Intake | crates/oris-intake |
Implemented baseline for issue intake/prioritization | intake-experimental |
| Evolution Network | crates/oris-evolution-network |
Experimental protocol scaffold | evolution-network-experimental |
| Full Framework Facade | crates/oris-runtime re-exports |
Aggregate framework surface | full-evolution-experimental |
The runtime layer is a supporting integration surface for hosting and operating the framework (execution server, workers, durable jobs). This README does not act as a runtime handbook; use the docs below when you need runtime-level details.
- EvoKernel docs index
- Evolution example suite
- Production operations guide
- Evo example programs
- EvoKernel overview
- License: MIT
- Attribution: This project includes code derived from langchain-rust.
- Contribution guide: CONTRIBUTING.md
- Code of conduct: CODE_OF_CONDUCT.md
- Security policy: SECURITY.md
- Privacy notice: PRIVACY.md
- Support guide: SUPPORT.md
- Governance: GOVERNANCE.md
- Crate: crates.io/oris-runtime
- API docs: docs.rs/oris-runtime
- Repository: GitHub