allowed_toolsstring[]
Tool names actually advertised to the route model. In route mode, well-formed calls outside this registry are suppressed before return. Defaults to Prism's seven trained routing tools.
cloud_fallbackboolean
If true, fall through to synalux portal cascade on local fail. Default false — token-saving mode is the point of this tool.
context_depthenum
Project-memory depth. Defaults to the Prism dashboard setting when project is provided.
conversation_idstring
Conversation id returned by session_bootstrap. Used for inference telemetry and continuity.
escalationenum
Failure contract (plan v2 §5.2). 'serve' (default) keeps legacy behavior: safety refusals throw, gate-failed output may be served. 'report' returns a structured gate_outcome on every terminal path — refused results come back as {status:'refused', output:''} instead of an error, and degraded (gate-failed, served-anyway) output is explicitly flagged so callers can distinguish success / degraded / refused.
evidenceobject[]
Optional evidence snippets the model output must be grounded in. When supplied with verify: true, every assertive claim in the draft (numbers, names, dates, codes, $ amounts) must be ENTAILED by one of these snippets or the draft is refused.
imagesstring[]
Screenshots or frames to analyse. Each entry is an absolute file path or raw base64. Requires a vision-capable tier; tiers without vision are skipped rather than shown the prompt without the image.
max_tokensnumber
Max output tokens (default 1024, hard cap 8192).
modeenum
Execution mode. 'route' (default) for MCP tool routing — fast, nothink. 'chat' for general conversation — uses thinking, escalates to cloud on failure. 'code' for code generation — uses thinking, larger context. In chat/code modes, prefers the 27B tier and enables <think> reasoning.
model_ceilingenum
Cap the largest tier the picker may select. e.g. '9b' forbids 27B even if RAM allows.
projectstring
Optional Prism project whose dashboard-depth handoff and recent session memory should be supplied to the local worker as historical data.
promptstringобязательный
The user prompt. Required.
route_guardenum
Route-output guard. 'auto' (default) applies the local advertised-tool contract and, for authenticated paid plans, the private Synalux deterministic route correction. 'local' keeps the prompt and draft entirely on-device.
strict_entitlementsboolean
Fail loud instead of running with ASSUMED free-tier limits (plan v2 §5.5). When true and entitlement resolution fell back to free because the portal was unreachable (source='fallback_free'), the call throws instead of silently applying free clamps. Portal-confirmed free plans and unconfigured machines are unaffected. Default: false.
systemstring
Optional system instruction prepended to the prompt.
task_complexitynumber
Optional deterministic 1-10 workload hint. prism_infer—not the task router—uses it to choose the initial local tier and thinking mode. Explicit model_ceiling/think overrides win.
temperaturenumber
Sampling temperature, 0 = deterministic (default 0).
thinkboolean
Enable thinking mode (<think> blocks). Default: true for chat/code, false for route. Thinking improves quality on complex tasks but adds latency (~2-5s).
timeout_msnumber
Override per-call timeout. Default scales with model size: 27B=120s, 9B=60s, 4B=20s, 2B=15s.
verifier_modelstring
Override the verifier model. Default: qwen3.5:4b.
verifier_timeout_msnumber
Override the verifier hard timeout. Default 2000 ms.
verifyboolean
Enable the L3 grounding verifier. Default: true when evidence is provided, false otherwise. When enabled, the model's draft is checked by a different model (qwen3.5:4b by default) against the supplied evidence. Drafts with NEUTRAL or CONTRADICTED claims are refused.
