Feature coverage map
This table is a quick index of the manual. “Direct CLI” means scientific-agent or scientific-agent-bench; capabilities that require serve plus curl are marked under API. Desktop and API use the same control plane, so core run facts agree even when interaction differs.
| Capability | Desktop | Direct CLI | Control API | Guide |
|---|---|---|---|---|
| First-run welcome and storage selection | ✓ | Global path options | storage endpoints | Desktop quickstart |
| Provider setup and connection test | ✓ | Environment + YAML | provider test | Settings |
| Codex CLI login detection | ✓ | codex login |
model profiles/provider test | Settings |
| Single-model and role-separated routing | ✓ | YAML | Frozen during run creation | Configuration |
| Browse capabilities and plugins | ✓ | Config directories/aliases | capabilities/plugins | Plugin use |
| Install/enable/disable directory plugin | ✓ | Direct config loading | plugin lifecycle | Plugin use |
| Install custom ZIP | ✓ | Extract and use directory | custom install | Plugin tutorial |
| Goal and research questions | Creation wizard | YAML/JSON | preview/create | Create a run |
| Multiple criteria and trusted metrics | Wizard/update before running | YAML | acceptance update | Configuration |
| Limits and remaining resources | Wizard/Control Room | YAML/status | budget/resources | Control Room |
| Capability allowlist and constraints | Wizard | YAML | preview/create | Configuration |
| Up to 20 input files | File picker | Upload through API | create/upload | Create a run |
| Scientific verification switch | Wizard | YAML | create | Scientific workflow |
| Artifact-completeness contract | Wizard | YAML | create | Scientific workflow |
| Frozen final predictor | Wizard | YAML | create | Configuration |
| Preview research contract | ✓ | Validated on run | /runs/preview |
Create a run |
| Create without starting | Create, then Start later | new |
create | CLI commands |
| Create and run | ✓ | run |
create + start | Quickstart |
| Pause and resume | ✓ | resume; pause via API |
pause/resume | Control Room |
| Cancel | ✓ | Through API | cancel | Control API |
| Interrupted recovery | ✓ | resume |
recover | Data and recovery |
| State and full events | ✓ | status --events |
state/events/SSE | CLI commands |
| Timeline filtering and details | ✓ | Event text | events | Control Room |
| State/Knowledge/Verification | Inspector | State summary | scientific queries | Scientific workflow |
| Reports | Inspector | Artifacts/state | reports | Scientific workflow |
| Model calls | Inspector | Status summary | model-calls | Control Room |
| Metrics/Resources/Budget/Jobs | Inspector | Status summary | Matching endpoints | Control Room |
| Ask about recorded research | ✓ | Through API | ask/interactions | Control Room |
| Add a research note | ✓ | Through API | intervene | Control API |
| Controlled budget update | ✓ | Through API | intervene | Control API |
| Update criteria before running | ✓ | Through API | acceptance | Control Room |
| Explicit scientific commands | Related UI actions | Through API | scientific-commands | Scientific workflow |
| Analyses, evidence, claims, hypotheses | Inspector/Lineage | Through API | Dedicated query endpoints | Scientific workflow |
| Verification plan, approval, execution, result | Inspector/commands | Through API | scientific commands/queries | Scientific workflow |
| Trusted evaluation/private-target isolation | State/results visible | Configure plugins | Trusted metrics queryable | Scientific workflow |
| Research Flow | ✓ | Through API | research-flow | Lineage |
| Full Provenance | ✓ | Through API | lineage | Lineage |
| Text/CSV artifact preview and download | ✓ | API/disk inspection | artifact content | Lineage |
| Lineage JSON/DOT/SVG/PNG export | ✓ | Process API data | lineage data | Lineage |
| Scientific result export | ✓ | Through API | exports | Data and recovery |
| Full Runtime backup | ✓ | Through API | storage backups | Data and recovery |
| View/change data location | ✓ | Global path options | storage info/location | Data and recovery |
| Diagnostics and logs | About | stderr/status | diagnostics | Troubleshooting |
| Update check | About (requires a configured release source) | Release channel | — | Settings |
| Repeated isolated trials | — | Benchmark run |
— | Benchmarks |
| Generalization evaluation | — | Benchmark generalization |
— | Benchmarks |
| Difficulty/replication evaluation | — | Benchmark difficulty |
— | Benchmarks |
| Develop Domain/Core plugins | Install/validate result | SDK/test/package | Install/discover | Plugin SDK |
| Develop an Executor | — | SDK/test | Use after discovery | Plugin SDK |
Current product boundaries
To distinguish an intentional boundary from missing documentation, the current release does not provide:
- multi-agent debate, voting, or learned model routing;
- ordinary Runtime/UI access to benchmark-private data;
- a Desktop benchmark dashboard;
- built-in Slurm/HPC or general remote execution service;
- a security sandbox for malicious plugins or generated code;
- automatic installation of third-party plugin dependencies;
- a promise of automatic theory discovery or general AutoML;
- arbitrary edits to historical events after a run has begun or succeeded.
These boundaries do not prevent new domain tools from being connected through plugins or automation from being built with the Control API inside a trusted network boundary.