Branch mode · next message will fork from the selected answer
240 ms
Search Scope
Selected Model · OpenRouter
AI Engine & API Connection
Vault status loading…
Storage: ciphertext only when enabledSession: decrypted key is never written to localStorage/sessionStorage
Telemetry & Cost Analytics
—TTFB / proxy TTFT
—Tokens / sec
—Total tokens
$0.0000Est. API cost
Provider usage is preferred; otherwise token counts use a character heuristic. Non-streaming endpoints expose TTFB as a proxy for TTFT.
Forest AI TelemetryCurrent session · active branch
Last Request—
—TTFB / TTFT proxyTime until response headers arrive.
—Total latencyFull agent loop duration.
—Output TPSCompletion tokens per second.
$0.0000Estimated costBased on selected model pricing matrix.
—Input tokensProvider usage or heuristic.
—Output tokensAcross all agent rounds.
0Tool callsClient-side agent calls this turn.
1Agent roundsLLM round trips before final answer.
Session Analytics0 requests
$0.0000Session cost
0Total tokens
—Avg latency
0%Error rate
Latency Trendlatest 24 requests
Recent Requests0 tool calls
Time
Model
Branch
Latency
TTFB
Tokens
TPS
Tools
Cost
Status
TTFT is exact only when the provider exposes token streaming. This prototype currently uses response-header arrival (TTFB) as the proxy on non-streaming calls, avoiding a falsely precise TTFT claim.
!
Approval RequiredForest AI wants to execute a sensitive browser-side tool.
PDF and DOCX parsing uses PDF.js and Mammoth inside an isolated Blob Web Worker. In this single-file prototype those parser libraries are loaded from a CDN at runtime; a production packaged build should pin and bundle them locally. Images are resized in the worker with createImageBitmap + OffscreenCanvas and encoded to WebP before storage/context use.