Terms of Service

1. Agreement to Terms

By downloading, installing, auditing, running, or accessing SiftrCode (including the siftrcode npm package, binary distributions, MCP servers, and the siftrcode.com website), you agree to be bound by these Terms of Service. If you do not agree to these terms, do not install or use our software.

2. Open Source License (Community CLI)

The core SiftrCode Command Line Interface and AST skeletonization engine are released under the terms of the MIT License.

You are free to audit, run locally in proprietary commercial projects, integrate into CI/CD pipelines, and contribute upstream via our GitHub repository.

3. Commercial Services (SiftrCode Pro)

For engineering teams utilizing SiftrCode Pro, team cross-repo context caching, or priority support:

• 14-Day Free Trial: Team trials are activated without upfront credit card requirements. At the end of the 14-day trial period, continuing service requires active billing enrollment.
• Seat Licensing: Pro licenses are billed per active engineering seat per month ($19/seat/mo). You may add or remove seats at any billing cycle boundary.
• Cancellations: You may cancel Pro subscriptions at any time via your dashboard or by emailing [email protected]. Service continues until the end of the paid billing interval.

4. Software Modifications & AI Agent Code Generation

SiftrCode provides AST interface skeletons, variable-resolution context bundles, and MCP telemetry tools to AI coding agents (Claude Code, Cursor, Antigravity, etc.) to optimize token windows. All parsing and optimization occurs locally on your machine with zero persistent raw source code retention. You acknowledge that software written or modified by autonomous AI agents remains the sole responsibility of your engineering team. SiftrCode does not guarantee that third-party LLMs will write bug-free code or adhere to requirements. You remain solely responsible for compiling, testing, and verifying all software changes before deploying to production.

5. Zero-Egress Technical Guarantees & Codebase Ownership

You retain 100% exclusive ownership of all codebases, files, repositories, syntax trees, and derivative outputs processed by SiftrCode. SiftrCode claims zero intellectual property rights, licenses, or title to your code:

• Local-First Guarantee: Candidate recall, ranking, knapsack budgeting, and discrete AST slicing execute entirely in local process memory on your machine.
• Authoritative Retention Rights: operationRights.*.retention.local is authoritative across every durable write; proprietary source code, snippets, and prompts are never retained in SQLite under default rights.
• Fail-Closed Remote Processing & Strict Gateway: Remote processing is denied by default. Any permitted external provider invocations are strictly governed by our EnforcedEgressGateway, which automatically sanitizes credentials (AWS keys, Stripe keys, Bearer tokens).
• Canonical Agent Lineage (Migration 13): Every local session, candidate decision, and optimization plan is strictly bound to a canonical AgentEnvironment.systemConfigurationHash identity, ensuring complete end-to-end environment traceability and auditability.
• Sanctioned Training Exporter Boundary: Direct persistence or querying of raw observations for training is strictly blocked by runtime policy. Only authenticated TrainingExporter exports carrying valid signed export tokens (texport_ / texport_ev_) and passing rights filters may generate persistent training rows.
• Post-Render Budget Degradation: When context limits are reached, the engine degrades AST units through an explicit ladder (FULL ➔ BODY ➔ SKELETON ➔ SIGNATURE ➔ NAME ➔ OMIT), with class method BODY isolation ensuring only relevant methods are provided while excluding sibling methods.
• No Model Training: Under no circumstances does SiftrCode utilize your code, prompts, or AST representations to train or tune machine learning models (trainingAllowed: false).

6. Benchmark Integrity & Shadow Judged Evidence Validation (JEV)

For empirical validation (including SiftrBench studies across 25 audited real-world tasks with 500 bounded shadow calls on Express, FastAPI, and SiftrCode, and the controlled Live JEV Smoke Study across 5 tasks with 25 bounded calls), Judged Evidence Validation (JEV) operates strictly under shadow invariants:

• 100% Plan Invariance: Shadow JEV calls never alter production context bundles, rank orders, or budget boundaries. Live smoke verification confirmed 100% normalized decision plan invariance (SHA-256 matched across units, resolutions, allocations, and exposures with zero deviation) across all runs. ContextRank remains the sole deterministic production decision-maker.
• Controlled Smoke Latency & Lineage: Live remote JEV calls operate within a bounded concurrency profile (peak concurrency 4) with P50 179ms latency (P95 446ms), zero database lineage mismatches, zero orphan signals, and authoritative WorkspaceSnapshot identity binding.
• Air-Gapped Telemetry: Benchmark telemetry records only structural metrics (AST unit counts, token latencies, continuous probability distributions) and never exposes underlying source code strings.

7. Disclaimer of Warranties

TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, SIFTRCODE IS PROVIDED "AS IS" AND "AS AVAILABLE", WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT.

8. Limitation of Liability

In no event shall SiftrCode, its contributors, authors, or corporate sponsors be liable for any indirect, incidental, special, consequential, or punitive damages, including loss of profits, data corruption, codebase disruption, or API token overages arising out of the use or inability to use the software.

9. Contact & Questions

If you have any questions regarding these Terms or need enterprise MSA licensing, reach out to: