Experience Synthetic Time Travel — type any moment in history, get a complete historically grounded scene in seconds: characters with distinct voices, period-accurate dialog, relationship dynamics, and a photorealistic image — all verified against Google Search.
curl -X POST localhost:8000/api/v1/timepoints/generate/stream \
-H "Content-Type: application/json" \
-d '{"query": "AlphaGo plays Move 37 against Lee Sedol in Game 2, Four Seasons Hotel Seoul March 10 2016", "generate_image": true}'Generated image:
Scene:
Location: Four Seasons Hotel, Seoul, South Korea
Date: March 10, 2016 — afternoon, spring
Tension: high
Setting: A brightly lit, modern conference room arranged for a Go tournament.
Central Go board on a low, polished wooden table. Lee Sedol sits
opposite the AlphaGo system, represented by a monitor displaying its
moves. Journalists, AI researchers, and Go enthusiasts seated in rows
behind a low barrier. Cables snake across the floor.
Atmosphere: Intense concentration and quiet anticipation. The audience is hushed.
A palpable sense of history being made, mixed with disbelief at
AlphaGo's unconventional moves.
Characters:
| Name | Role | Voice |
|---|---|---|
| Lee Sedol | primary | Short, halting fragments — stunned understatement |
| AlphaGo | primary, silent | Represented by monitor (non-human entity resolved by grounding agent) |
| Commentator 1 | secondary | Chatty, analytical, comfortable on-air cadence |
| Commentator 2 | secondary | More measured, technical observations |
| Tournament Official | secondary | Formal, procedural |
Dialog:
Lee Sedol: "...Huh. That's... certainly a move."
Commentator 1: "Well, that's one way to put it. I'm not sure I've ever seen a professional play quite like that before."
Lee Sedol: "It's... unexpected, to say the least. Maybe it's a bluff?"
Commentator 1: "A bluff? Against AlphaGo? Now that's a thought."
Commentator 2: "It's a bold one, if so. I'm not sure even a human could read that."
Commentator 1: "It's either genius or madness, and I honestly can't tell which right now."
Lee Sedol: "Perhaps both, a little? It's making things... interesting."
Then interrogate the characters:
> You: "Lee Sedol, what went through your mind when you saw Move 37?"
Or jump forward in time:
POST /api/v1/temporal/{id}/next {"units": 1, "unit": "hour"}
# → One hour later: Lee Sedol has left the room. The commentators are still trying to explain it.Prerequisites: Python 3.10+ and a Google API key (free at AI Studio)
git clone https://github.com/timepointai/timepoint-flash.git
cd timepoint-flash
./setup.sh # Checks prereqs, installs deps, creates .env
# Edit .env → add your GOOGLE_API_KEY
./quickstart.sh # Starts server + generates a demo sceneOr manually:
pip install -e .
cp .env.example .env # Add your API key
./run.sh -r # Start server
./demo.sh # Interactive demo with 10 templatesSwagger docs at http://localhost:8000/docs
Import the repo directly from GitHub — Replit reads .replit and runs automatically.
- Create a new Replit → Import from GitHub →
https://github.com/timepointai/timepoint-flash - Set Secrets (sidebar → Secrets tab):
GOOGLE_API_KEY— get free at aistudio.google.comOPENROUTER_API_KEY— get at openrouter.ai (optional, enables hyper/gemini3 presets)
- Hit Run — the server starts on port 8080 and Replit exposes a public URL
The startup script (start.sh) handles dependency installation, database migrations, and schema patching automatically on every run. No manual setup required.
Swagger docs at https://your-repl.replit.dev/docs
14 specialized agents run a pipeline with parallel execution, Google Search grounding, and a critique-retry loop:
Judge → Timeline → Grounding (Google Search) → Scene
↓
Characters (grounded) + Moment + Camera [parallel]
↓
Dialog → Critique (auto-retry if issues)
↓
ImagePrompt → Optimizer → ImageGen
Key capabilities:
- Google Search grounding — Verified locations, dates, participants. Not "a room in New York" but "35th floor of the Equitable Center, Manhattan."
- Critique loop — Dialog is reviewed for anachronisms, cultural errors (Greek vs Roman deities), modern idioms, and voice distinctiveness. Auto-retries with corrections if critical issues found.
- Voice differentiation — Each character gets a social register (elite/educated/common/servant/child) that constrains sentence structure, vocabulary, and verbal tics. Characters must be identifiable by voice alone.
- Emotional transfer — The image prompt optimizer translates narrative tension into physicalized body language instead of discarding it. "Climactic tension" becomes "wide eyes, dropped objects, body recoiling."
- Entity representation — Non-human entities (Deep Blue, AlphaGo, HAL 9000) are shown through their physical representatives (IBM operator, monitor display, red camera lens).
- Anachronism prevention — Era-specific exclusion lists, mutual exclusion rules (Roman toga + tricorn hat), famous painting drift detection.
- 3-tier image fallback — Google Imagen → OpenRouter Flux → Pollinations.ai. Image generation never fails.
The Clockchain Connection — Flash is the Reality Writer for the Clockchain. Every scene becomes a verified anchor point in the temporal causal graph, strengthening the Bayesian prior for all future renderings. Flash's Google Search grounding ensures historical accuracy today; as the Clockchain accumulates validated causal paths, it becomes a complementary grounding source. Flash scenes can be exported as TDF records for use across the Timepoint suite.
Oppenheimer at Trinity, 5:29 AM July 16 1945:
Location: Control bunker S-10000, Jornada del Muerto desert, New Mexico
Characters: J. Robert Oppenheimer (director), Leslie Groves (general),
Kenneth Bainbridge (test director), Enrico Fermi (physicist)
Dialog:
Bainbridge: "Minus ten seconds."
Fermi: [tearing strips of paper, holding them above his head to measure blast yield]
Oppenheimer: "..."
Bainbridge: "Now we are all sons of bitches."
Gavrilo Princip at Schiller's Deli, Sarajevo 1914:
Location: Moritz Schiller's Delicatessen, Latin Bridge, Sarajevo
Date: June 28, 1914 — late morning
Tension: extreme — the car took a wrong turn and stopped right in front of him
Characters: Gavrilo Princip (19-year-old), Archduke Franz Ferdinand,
Sophie Chotek, Driver Leopold Lojka
Ramanujan arrives at Trinity College, March 1914:
Location: Great Gate, Trinity College, Cambridge
Date: March 1914 — freezing rain, late afternoon
Characters: Srinivasa Ramanujan (26, mathematician), College Porter
Dialog:
Porter: "Name?"
Ramanujan: "Ramanujan. I have a letter from Mr. Hardy."
Porter: "Everyone's got a letter from someone, haven't they."
Each scene includes full character bios, relationship graphs, scene metadata, camera composition, and a generated image.
| Preset | Speed | Provider | Best For |
|---|---|---|---|
| hyper | ~55s | OpenRouter | Fast iteration, prototyping |
| balanced | ~90-110s | Google Native | Production quality |
| hd | ~2-2.5 min | Google Native | Maximum fidelity (extended thinking) |
| gemini3 | ~60s | OpenRouter | Latest model, agentic workflows |
# Hyper for speed
curl -X POST localhost:8000/api/v1/timepoints/generate/stream \
-d '{"query": "Kasparov Deep Blue Game 6 Equitable Center 1997", "preset": "hyper", "generate_image": true}'
# HD for quality
curl -X POST localhost:8000/api/v1/timepoints/generate/stream \
-d '{"query": "Oppenheimer Trinity test 5:29 AM July 16 1945", "preset": "hd", "generate_image": true}'Downstream apps can control model selection and generation behavior per-request:
model_policy: "permissive"— Routes all generation through open-weight models (DeepSeek, Llama, Qwen, Mistral) via OpenRouter, uses Pollinations for images, and skips Google grounding. Fully Google-free.text_model/image_model— Override preset models with any OpenRouter-compatible model ID (e.g.qwen/qwen3-235b-a22b) or Google native model.llm_params— Fine-grained control over temperature, max_tokens, top_p, top_k, penalties, stop sequences, and system prompt injection. Applied to all 14 pipeline agents.
All three are composable: model_policy + explicit models + llm_params work together with clear priority ordering. See docs/API.md for the full parameter reference.
| Endpoint | Description |
|---|---|
POST /api/v1/timepoints/generate/stream |
Generate scene with SSE progress (recommended) |
POST /api/v1/timepoints/generate/sync |
Generate scene, block until complete |
POST /api/v1/timepoints/generate |
Background generation, poll for result |
GET /api/v1/timepoints/{id} |
Retrieve a completed scene |
PATCH /api/v1/timepoints/{id}/visibility |
Set public/private visibility |
POST /api/v1/interactions/{id}/chat |
Chat with a character |
POST /api/v1/temporal/{id}/next |
Jump forward in time |
POST /api/v1/temporal/{id}/prior |
Jump backward in time |
GET /api/v1/temporal/{id}/sequence |
Get linked timeline |
POST /api/v1/auth/apple |
Apple Sign-In → JWT pair |
POST /api/v1/auth/refresh |
Rotate refresh token |
GET /api/v1/credits/balance |
Credit balance |
GET /api/v1/credits/costs |
Credit cost table |
GET /api/v1/users/me/timepoints |
User's scene history |
POST /api/v1/users/resolve |
Find or create user by external_id (service-key protected) |
POST /api/v1/eval/compare |
Compare model latencies |
GET /api/v1/models/free |
List free OpenRouter models |
Full reference: docs/API.md
# .env
GOOGLE_API_KEY=your-key # Required (free at aistudio.google.com)
OPENROUTER_API_KEY=your-key # Optional (for hyper/gemini3 presets)
DATABASE_URL=sqlite+aiosqlite:///./timepoint.db # Default storage
# Auth (optional — for iOS app mode)
AUTH_ENABLED=false # Set true to require JWT auth
JWT_SECRET_KEY=change-me-to-a-random-32-char-string # Required when AUTH_ENABLED=true
APPLE_BUNDLE_ID=com.yourcompany.app # Required when AUTH_ENABLED=true
SIGNUP_CREDITS=50 # Free credits on first sign-in
SHARE_URL_BASE=https://timepointai.com/t # Optional: enables share_url in responses
# Service-to-service auth (optional — for multi-service deployment)
FLASH_SERVICE_KEY= # Shared secret for X-Service-Key header. Empty = disabled.
CORS_ENABLED=true # Set false when Flash is internal-only (no browser callers)
# Billing (optional — see app/services/billing.py)
BILLING_ENABLED=false # Set true to enable billing integrationThe open-source app ships with NoOpBilling — all credit checks return true (unlimited access). The BillingProvider protocol in app/services/billing.py defines hooks for custom billing integrations. The deployed version uses a separate billing microservice for Apple IAP and Stripe payments.
See .env.example for the full list of environment variables.
python3.10 -m pytest tests/ -v # 522 fast + integration, 11 skipped (e2e needs API keys)660+ tests covering generation, character interactions, temporal navigation, image fallback, historical grounding, schema validation, visibility/access control, and provider failover.
- API Reference — Full endpoint documentation
- iOS Integration — Auth flow, credit system, endpoint map for iOS client
- Agent Architecture — Pipeline breakdown with example output
- Temporal Navigation — Time travel mechanics
- Downstream Model Control — Model policy, LLM params, and per-request control for downstream apps
- Eval Roadmap — Quality scoring and benchmark plans
- Deployment — Local, Replit, and production deployment
Open-source engines for temporal AI. Render the past. Simulate the future. Score the predictions. Accumulate the graph.
| Service | Type | Repo | Role |
|---|---|---|---|
| Flash | Open Source | timepoint-flash | Reality Writer — renders grounded historical moments (Synthetic Time Travel) |
| Pro | Open Source | timepoint-pro | Rendering Engine — SNAG-powered simulation, TDF output, training data |
| Clockchain | Open Source | timepoint-clockchain | Temporal Causal Graph — Rendered Past + Rendered Future, growing 24/7 |
| SNAG Bench | Open Source | timepoint-snag-bench | Quality Certifier — measures Causal Resolution across renderings |
| Proteus | Open Source | proteus | Settlement Layer — prediction markets that validate Rendered Futures |
| TDF | Open Source | timepoint-tdf | Data Format — JSON-LD interchange across all services |
| Web App | — | — | Browser client at app.timepointai.com |
| iPhone App | — | — | iOS client — Synthetic Time Travel on mobile |
| Billing | — | — | Payment processing — Apple IAP + Stripe |
| Landing | — | — | Marketing site at timepointai.com |
Apache 2.0
Built with Python, FastAPI, and Google Gemini. Part of Timepoint AI.