Evaluate questions over supplied evidence
POST /v1/decide
MCP tool: sysone_decideAsk up to eight questions over one text input of at most 12,000 characters. Boolean questions return a probability. Choice questions select from named options. Score questions use your ordered rubric.
Use it to check a claim, rate candidate context or recommend the next tool. Include the actual evidence and define what uncertainty should cause the caller to do. The result cannot establish a fact missing from the input.
Missing probability stays unknown. The caller chooses its review threshold. A valid answer format does not prove the answer is correct.
Triage records for further inspection
POST /v1/logs
MCP tool: sysone_logsThe jevlogs service accepts up to 16 records, each with an ID, body and optional severity or protection flag. It helps decide which records need further analysis.
Keep original records in your logging system. Protected errors and uncertain results remain eligible for inspection. A routine classification is advice about analysis priority, not permission to delete a record.
Our public log study found that a conservative policy sent almost everything for deeper analysis and increased the modeled bill. Compare severity rules and exact reuse before adding inference.
Select a category in a hierarchy
POST /v1/tree
MCP tool: sysone_treeThe jev-tree service takes a question, evidence and your category hierarchy. Set a call budget between one and 16. The default is eight. Ambiguous paths return a signal for review.
A hierarchy can reach categories that a truncated flat list excludes. It also adds calls. When a structured field already names the category, parse that field instead.
Recommend a recorded cue or silence
POST /v1/dialogue
MCP tool: sysone_dialogueThe application supplies a context ID, room, trigger, timing, completed cues and up to eight eligible recordings. System One can recommend one cue, silence or a deterministic fallback.
The Infinite Parable integration
The game filters completed cues and enforces spacing before asking for advice. It rejects a response when its context ID no longer matches the active scene. The narration director owns playback, interruption and continuation.
The optional service chooses among authored recordings. It does not write a script, synthesize speech or grant game progress. There is no independent playthrough study establishing that it improves the experience yet.
Discover recipes before making a call
The MCP connection has seven tools. Use discovery to find a recipe, then run it with your evidence. The four direct service tools remain available for custom inputs.
| Tool | Purpose |
|---|---|
sysone_status | Read service availability, connection limits and engine status without inference. |
sysone_patterns | List recipes or fetch one recipe by ID with its editable input and evidence link. |
sysone_run | Run a decision recipe with state and optional candidate descriptions. |
sysone_decide | Evaluate typed questions over supplied evidence. |
sysone_logs | Triage a bounded set of log records. |
sysone_tree | Select a category within a supplied hierarchy. |
sysone_dialogue | Advise on eligible recorded dialogue. |
A credential determines the available services. Naming a client does not grant permission. See the MCP connection instructions.
Control usage at the connection
All services use scoped credentials, daily request and model-call limits, bounded concurrency and cancellation. Exact response reuse stays within the connection. Failed model attempts remain counted.
Studio stores usage metadata such as tokens, latency and outcome. It does not store request bodies in the usage ledger. Read what the app sends and stores.
Use the SDK or HTTP API from application code. The public package exports request schemas, and Studio lets you edit and inspect each service input.