AI for Dart and Flutter Production Apps
The key to Dartastic Hosted Observatory magic is bringing together your packaged code published to Dartastic Pub and OpenTelemetry from your live users. Dartastic AI runs on the same Dartastic Hosted Observatory stack and can detect problems as they happen and detect and produce fixes as problems arise instead of waiting for humans to process tickets. Instead of getting problems, you get solutions.
We don't train on your data
Your traces, logs, and source code are never used to train. Inference reads your data; training doesn't. Your security and our integrity are our primary design forces, not profit.
Runs where your data lives
Dartastic AI runs on Dartastic Hosted Observatory stacks.
Or bring your own key
Prefer Claude, GPT, or Bedrock? BYOK lets you point the Dartastic agent layer at your own provider with your own zero-retention contract in place. We never see the key or the responses.
The structural difference
Every observability AI on the market today is multi-tenant cloud. Your queries, your traces, your stack frames travel to a vendor's infrastructure and sit alongside every other customer's data. Some vendors promise they won't train on yours; some have been caught doing it anyway. Every privacy-conscious org's legal team eventually flags this.
Dartastic is single-tenant by design. Every customer gets a dedicated box. Add Dartastic AI and that pattern extends to the model: your data never crosses a vendor boundary, period. Inference happens either against our central AI box (a Dartastic-owned service trained on Dartastic data) or on your own box for the air-gapped tiers. No competitor can match this without rebuilding their multi-tenant architecture.
What it does
For mobile and Flutter developers
- "Why did this trace drop frames?" — grounded in the jank-attribution span and the symbolicated source line that caused it.
- "What regressed between v1.4 and v1.5?" — compares release-health snapshots, surfaces statistically-significant deltas.
- PR review against your existing telemetry: "this widget rebuild will regress jank by ~20%."
- Test generation in your project's existing style.
For server and Dart-server developers
- Trace explanation with source-line citations. Full source is on the box via Dartastic Pub.
- "Tune this sampler" — analyzes actual cardinality + cost, recommends a config.
- "Migrate from Sentry" — leverages the existing envelope translator.
For DevOps and on-call
- Incident triage grounded in your history, not a generic LLM guess.
- Cost optimization — cardinality audits, log-volume hot spots, sampler recommendations.
- Capacity planning from the box's own resource trends.
For everyone
- Every answer cites a span ID, source line, or metric. Refuses to answer without a citation.
- No hallucinated APIs — RAG is grounded in your on-box Dartastic Pub packages.
- Stickiness compounds: the AI gets better at your codebase over time.
Pricing
Shared AI
$50/mo
Limited-usage inference against the central Dartastic AI box. Authenticated HTTPS from your Hosted box. Best for getting started.
Private AI on Box
$100/mo
Same model, deployed on your own Hosted box. Higher usage cap. No cross-box network calls. Requires a Hosted Pro+ tier.
BYOK
$25 – $50/mo
Bring your own Anthropic, OpenAI, or Bedrock key. Dartastic orchestrates; we never see the key or the response. Orthogonal — combine with any tier. Free on Compliance runtime.
AI add-ons require a Dartastic Hosted subscription as the base.
For regulated environments — Compliance runtime
Compliance is a runtime tier, not a Hosted add-on. It sits at the top of the runtime ladder above the Pro runtime, and is the only way to get fully air-gapped AI on a customer-controlled box.
Compliance runtime customers get the locked-binary build (no outbound network calls except to their own infrastructure), the Dartastic-trained AI deployed on their own Hosted box (no central AI-box call, ever), BYOK orchestration included, and the attestation paperwork that SOC2 / HIPAA-adjacent / data-residency reviews need. Designed for defense, healthcare, finance, and regulated banks.
Compliance runtime
$999/mo
- Air-gapped Private AI on box
- No outbound LLM calls, ever
- BYOK included at no extra cost
- SOC2 / HIPAA-adjacent attestation
- Locked-binary build (#106/#107)
Which model
The Dartastic-trained base model is Llama or Mistral — open-weight, US/EU origin, license permits commercial fine-tuning. We deliberately do not ship Qwen, DeepSeek, or any other Chinese-origin weights; your customers' data should not transit a model that's subject to a foreign government's jurisdiction.
Anthropic Claude is excellent for inference but their terms prohibit training derivative models on Claude outputs — so Claude is available under BYOK, not as the base of our trained model. Same for GPT and Bedrock.
Honest about the limits
AI suggestions are advisory. Output from Dartastic AI is generated by a machine-learning system based on your telemetry and source code. It is provided for informational purposes only. You are solely responsible for reviewing, validating, testing, and accepting any code, configuration, or operational change suggested by Dartastic AI before applying it. Dartastic makes no warranty of accuracy, completeness, or fitness for purpose.
Training data attestation: Dartastic AI is trained on Dartastic-owned telemetry and public Dart/Flutter packages. Customer telemetry is never used to train the model. When using BYOK, your queries and context are sent directly from your Hosted box to your chosen LLM provider using credentials you supply; Dartastic does not intercept, log, or retain these requests. Under Compliance tier, the model runs entirely on infrastructure you control; no outbound LLM call leaves your box.