Repository navigation
Expand file tree
/
Copy path.env.example
More file actions
49 lines (42 loc) · 1.94 KB
/
Copy path.env.example
File metadata and controls
49 lines (42 loc) · 1.94 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
# Copy this file to .env.local (gitignored) and fill in real values.
# Loaded by the runtime CLIs:
# npm run warm (Tensorlake pre-warm before demo)
# npm run live-smoke (full SDK round-trip + gate path smoke)
# npm run build:phi-image (register the Phi-3-mini sandbox image)
# The Tensorlake SDK auto-reads TENSORLAKE_* from process.env, so the runtime
# clients pick these up without further wiring.
# Tensorlake — required for the live demo path. Without these set, the runtime
# silently falls back to LocalFakeTensorlakeClient (still works offline; no
# real sandbox spawns visible in dashboard).
TENSORLAKE_API_KEY=
TENSORLAKE_ORGANIZATION_ID=
TENSORLAKE_PROJECT_ID=
# Optional: override default https://api.tensorlake.ai
# TENSORLAKE_API_URL=
# Nia — Tier-1 vault for compiled functions. Daemon writes here on every fire.
NIA_API_KEY=
NIA_VAULT_ID=
# Tensorlake image registered with `npm run build:phi-image`. Must match the
# image name in build-phi-image.ts (default: compile-phi-mini). When unset,
# RealTensorlakeClient.runPhi throws → TensorlakeWithLocalFallback engages.
COMPILE_PHI_IMAGE=compile-phi-mini
COMPILE_PHI_MODEL=phi3:mini
# Used by packages/synthesizer/harness — Friday derisk #1 only.
# Production MCP server does NOT use this; per D9, codegen happens in the
# customer's agent on their own LLM keys.
ANTHROPIC_API_KEY=
# Tensorlake (Lane B — 64-worker grid, Phi-3-mini hosting).
TENSORLAKE_API_KEY=
TENSORLAKE_ORGANIZATION_ID=
TENSORLAKE_PROJECT_ID=
# Nia (Lane D — Vault, Document Agent, semantic search).
NIA_API_KEY=
NIA_VAULT_ID=
# Convex (Lane C — reactive state for the eleven-page demo).
# Local dev: written automatically by `npx convex dev`.
# Production (Vercel): set CONVEX_DEPLOY_KEY in the Vercel project so the
# build runs `npx convex deploy` against the prod deployment, then expose
# CONVEX_URL to the runtime so the stream adapter and UI can subscribe.
CONVEX_DEPLOYMENT=
CONVEX_URL=
CONVEX_DEPLOY_KEY=