See Fluid Forge in action

Five short demos. Each one starts with the pain. Each one closes with a number you remember. Real captures, real numbers — every cast below is a frame-perfect SVG of the actual CLI output. Click play, read the takeaway banner, move on.

Pick your problem

Your pain	Watch this	Time	Punchline
Writing 200 lines of Python per data source	`$0.03` per data product	~50 s	Three providers. Same contract. Free on Ollama.
Six months to set up Airbyte for one Postgres	Six months → sixty seconds	~40 s	Six engines. One contract. Zero clusters.
Most CLIs ask 27 questions before they help	23 questions, skipped	~50 s	47 ms scan. 23 of 27 questions auto-resolved.
3am Slack ping: pipeline broke, who knows why	Skip the panic	~75 s	3am incident → ship in 90 seconds.
Your agent loop quietly spent `$0.50` on tool history	`$0.50` → `$0.05`	~40 s	One env var. 10× cheaper agent loops.

$0.03 per data product

Most data teams write 200 lines of Python per source. Or 8 lines of YAML and let the AI build the contract — same intent, three providers (Anthropic Haiku 4.5, OpenAI gpt-4.1-mini, local Ollama gemma2), real production cost figures on screen. Switch providers any time. The contract stays portable.

fluid forge data-model — same intent, three providers, real cost figures

Eight lines of intent YAML. Anthropic, OpenAI, Ollama — same flag swapped each time. All three emit the same valid contract, the same dbt project layout. Real production token counts and costs.

Pairs with Forge Data Model and LLM Providers. Long-form animated reel preserved at /reels/forge-in-action.html.

Six months → sixty seconds

Six months of Airbyte. Two weeks of Airflow DAGs. JVM heap tuning. For one Postgres source. Or fluid init --discover postgres://… and a 60-second flight to a working Bronze contract — switch engine: between duckdb, dlt, meltano, airbyte, kafka-connect, debezium when you outgrow embedded mode.

fluid init --discover postgres://... — Bronze contract in 60 seconds

Connect, scan information_schema (47 ms), infer 28 tables / 143 columns / 12 PII candidates / 8 foreign keys, emit a complete Bronze contract.fluid.yaml, validate, apply against embedded DuckDB. 6.2 seconds total. Then show the engine swap path.

Pairs with Source-Aligned Acquisition, Postgres → DuckDB walkthrough, and fluid init --discover. Long-form animated reel preserved at /reels/source-aligned-bronze.html.

23 questions, skipped

Most CLIs ask 27 questions before they help you. Forge asks four — the rest, it already knows. 47 ms welcome scan, 5-mode picker, inferences-first interview, slash commands at every prompt, cost-cap progress prefix in real time, and a pre-write panel so nothing surprises you.

fluid forge — guided UX in action

47 ms welcome scan finds 3 CSVs + 2 dbt models + 1 README, infers domain (finance) and PII (5 columns). 5-mode picker. 4 questions answered (most accept the inferred default with ↵). Cost-cap progress in real time ($0.000 → $0.021 of $0.050 cap). Pre-write panel shows exactly what will + won't change.

Pairs with Guided fluid forge UX and the fluid forge reference. Long-form animated reel preserved at /reels/guided-forge-ux.html.

Skip the panic

It's 3am. Pipeline broke. You have 90 seconds. fluid runs status (where), fluid runs logs --component dlq (why), fluid runs diff (what), one-line policy fix, fluid ship. Day-1 ships. Day-2 doesn't surprise.

3am Slack ping → ship in 90 seconds

PagerDuty: freshness SLA breached. fluid runs status shows 3 consecutive failures. fluid runs logs --component dlq surfaces the root cause (CHECK constraint NOT NULL, 47 partial-window customers). fluid runs diff shows what changed since the last OK run. One-line contract fix: NOT_NULL → NOT_NULL_WHERE customer_age_days >= 30. fluid ship. 87 seconds end-to-end. Recovered 12,361 rows.

Pairs with fluid runs, fluid retention, fluid secrets, fluid stats, and Typed CLI Errors. Long-form animated reel preserved at /reels/day2-ops.html.

$0.50 → $0.05

Long agent loops accumulate tool results — every turn rides on top of the last one. Without compaction, your bill grows super-linearly. Three strategies — truncate (free, default), summarize (LLM-backed, high recall), hybrid (cheap path first, summariser as safety net) — pick yours, set one env var, watch the bill drop 10×.

Agent-loop compaction — three strategies, real before/after costs

20-turn baseline: $0.503/run, super-linear context bloat (5K → 67K → 298K tokens). Three strategies side-by-side: truncate (5.8× cheaper), summarize (9.3× cheaper), hybrid (10.5× cheaper). One env var: FLUID_COMPACTION_STRATEGY=hybrid. No code changes.

Pairs with Agentic primitives → Token-budget pre-flight & compaction. Long-form animated reel preserved at /reels/compaction-and-warnings.html.

More demos in 30 seconds each

The full library — 14 frame-perfect SVG casts in all — covers AWS, Snowflake live-auth dry-run, agent-policy enforcement, blank scaffolds, policy compilation, and the canonical local quickstart:

→ Browse all demos

How these are sourced

Each cast above is a deterministic asciinema-rendered SVG generated by a Python pipeline (scripts/cast_builder.py → scrub-cast.py → svg-term). The narrative lives in scripts/demos/<name>.py — clone, edit, regenerate. Each cast plays inline on every browser without GIFs, JS frameworks, or third-party CDNs.

Token counts, durations, and costs in LLM-driven content (forge-multi-provider, agent-compaction) are real captures from production runs. Latency numbers in deterministic content (source-aligned-bronze, guided-forge-ux, day2-ops) are representative of the v0.8.3 stack against the bundled examples/ fixtures.

The original animated HTML reels are preserved at /reels/ for anyone who wants the longer-form pacing — each cast section above links through to its corresponding reel.