{"product":"a11oy Code","issuer":"a11oy","doctrine":"v11","tagline":"A governed agentic coder you can mathematically trust. NOT AGI.","modes":["chat","code","research"],"governed_loop":"Every turn runs through the PROVEN P1-P6 6-receipt loop (retrieve -> quarantine -> tool_call -> policy_check -> kernel_check -> emit) and emits a signed, re-verifiable receipt.","open_weight_roster":[{"model_id":"deepseek_coder_v2_instruct","display_name":"DeepSeek-Coder-V2-Instruct","provider":"DeepSeek","provider_slug":"deepseek","open_weight":true,"tier_band":"capable","hf_repo":"deepseek-ai/DeepSeek-Coder-V2-Instruct","params":"236B (MoE, 21B active)","context_window":128000,"license":"DeepSeek License v1.0 (code: MIT)","commercial_ok":true,"license_restrictions":["no military use","use-based restrictions (Attachment A)"],"redistribute_weights":false,"serving":"tower-side (GPU) — load by hf_repo at runtime; NOT served in CPU Space","code_strength":"GPT-4-Turbo-class on code (per DeepSeek model card claims; not re-benchmarked here)","use_case":"Primary coding brain — agentic code edits, multi-file PRs, debugging","router_role":"CODE_PRIMARY","modalities":["text","code"],"streaming":true,"tier_name":"alloy_capable","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; tower-side (GPU) — load by hf_repo at runtime; NOT served in CPU Space"},{"model_id":"deepseek_v2_5","display_name":"DeepSeek-V2.5","provider":"DeepSeek","provider_slug":"deepseek","open_weight":true,"tier_band":"capable","hf_repo":"deepseek-ai/DeepSeek-V2.5","params":"236B (MoE, 21B active)","context_window":128000,"license":"DeepSeek License v1.0","commercial_ok":true,"license_restrictions":["no military use","use-based restrictions (Attachment A)"],"redistribute_weights":false,"serving":"tower-side (GPU)","code_strength":"Strong general+code; merges chat+coder (DeepSeek card)","use_case":"General reasoning + code blend; orchestration fallback for DeepSeek-Coder-V2","router_role":"CODE_GENERAL","modalities":["text","code"],"streaming":true,"tier_name":"alloy_capable","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; tower-side (GPU)"},{"model_id":"qwen2_5_coder_32b_instruct","display_name":"Qwen2.5-Coder-32B-Instruct","provider":"Alibaba Qwen","provider_slug":"qwen","open_weight":true,"tier_band":"capable","hf_repo":"Qwen/Qwen2.5-Coder-32B-Instruct","params":"32.5B","context_window":128000,"license":"Apache-2.0","commercial_ok":true,"license_restrictions":[],"redistribute_weights":true,"serving":"tower-side (GPU)","code_strength":"GPT-4o-competitive on code benchmarks (Qwen card claims)","use_case":"Permissive (Apache) coding brain — preferred when license-clean weights matter","router_role":"CODE_PRIMARY_PERMISSIVE","modalities":["text","code"],"streaming":true,"tier_name":"alloy_capable","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; tower-side (GPU)"},{"model_id":"llama_3_3_70b_instruct","display_name":"Llama-3.3-70B-Instruct","provider":"Meta","provider_slug":"meta","open_weight":true,"tier_band":"capable","hf_repo":"meta-llama/Llama-3.3-70B-Instruct","params":"70B","context_window":128000,"license":"Llama 3.3 Community License","commercial_ok":true,"license_restrictions":["commercial OK only under 700M MAU","acceptable use policy","Built-with-Llama attribution"],"redistribute_weights":false,"serving":"tower-side (GPU)","code_strength":"Strong general + code; 405B-class quality at 70B (Meta card)","use_case":"General reasoning brain; broad-knowledge fallback","router_role":"GENERAL","modalities":["text","code"],"streaming":true,"tier_name":"alloy_capable","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; tower-side (GPU)"},{"model_id":"qwen2_5_coder_14b_instruct","display_name":"Qwen2.5-Coder-14B-Instruct","provider":"Alibaba Qwen","provider_slug":"qwen","open_weight":true,"tier_band":"mid","hf_repo":"Qwen/Qwen2.5-Coder-14B-Instruct","params":"14.7B","context_window":128000,"license":"Apache-2.0","commercial_ok":true,"license_restrictions":[],"redistribute_weights":true,"serving":"tower-side (GPU) or strong CPU/quantized","code_strength":"Very strong code at mid cost","use_case":"Cost-aware coding brain","router_role":"CODE_MID","modalities":["text","code"],"streaming":true,"tier_name":"alloy_mid","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; tower-side (GPU) or strong CPU/quantized"},{"model_id":"qwen2_5_coder_7b_instruct","display_name":"Qwen2.5-Coder-7B-Instruct","provider":"Alibaba Qwen","provider_slug":"qwen","open_weight":true,"tier_band":"mid","hf_repo":"Qwen/Qwen2.5-Coder-7B-Instruct","params":"7.6B","context_window":128000,"license":"Apache-2.0","commercial_ok":true,"license_restrictions":[],"redistribute_weights":true,"serving":"tower-side or quantized CPU (slow)","code_strength":"Strong code for size","use_case":"Mid coding brain","router_role":"CODE_MID","modalities":["text","code"],"streaming":true,"tier_name":"alloy_mid","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; tower-side or quantized CPU (slow)"},{"model_id":"deepseek_coder_6_7b_instruct","display_name":"DeepSeek-Coder-6.7B-Instruct","provider":"DeepSeek","provider_slug":"deepseek","open_weight":true,"tier_band":"mid","hf_repo":"deepseek-ai/deepseek-coder-6.7b-instruct","params":"6.7B","context_window":16000,"license":"DeepSeek License v1.0 (code: MIT)","commercial_ok":true,"license_restrictions":["no military use","use-based restrictions"],"redistribute_weights":false,"serving":"tower-side or quantized CPU (slow)","code_strength":"Strong code for size (DeepSeek card)","use_case":"Mid DeepSeek coding brain","router_role":"CODE_MID","modalities":["text","code"],"streaming":true,"tier_name":"alloy_mid","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; tower-side or quantized CPU (slow)"},{"model_id":"codestral_22b","display_name":"Codestral-22B","provider":"Mistral AI","provider_slug":"mistral","open_weight":true,"tier_band":"mid","hf_repo":"mistralai/Codestral-22B-v0.1","params":"22B","context_window":32000,"license":"Mistral AI Non-Production License (MNPL)","commercial_ok":false,"license_restrictions":["NON-PRODUCTION ONLY — research/eval/personal","no commercial use, no hosted/managed service","commercial license available from Mistral on request"],"redistribute_weights":false,"serving":"research/eval ONLY — EXCLUDED from production routing by default","code_strength":"Strong code (Mistral card)","use_case":"Benchmark/eval context only — NOT a production brain","router_role":"EXCLUDED_NONCOMMERCIAL","modalities":["text","code"],"streaming":true,"tier_name":"alloy_mid","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; research/eval ONLY — EXCLUDED from production routing by default"},{"model_id":"qwen2_5_coder_1_5b_instruct_gguf","display_name":"Qwen2.5-Coder-1.5B-Instruct (GGUF, llama.cpp)","provider":"Alibaba Qwen","provider_slug":"qwen","open_weight":true,"tier_band":"demo_cpu","hf_repo":"Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF","params":"1.54B","context_window":32000,"license":"Apache-2.0","commercial_ok":true,"license_restrictions":[],"redistribute_weights":true,"serving":"LIVE on CPU Space via llama.cpp IF weights present (slow, real); else honest tower-side label","code_strength":"Real small-coder output; not GPT-class but genuine","use_case":"LIVE in-Space demo brain (constrained CPU)","router_role":"DEMO_CPU","modalities":["text","code"],"streaming":false,"tier_name":"alloy_demo_cpu","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; LIVE on CPU Space via llama.cpp IF weights present (slow, real); else honest tower-side label"},{"model_id":"qwen2_5_coder_0_5b_instruct_gguf","display_name":"Qwen2.5-Coder-0.5B-Instruct (GGUF, llama.cpp)","provider":"Alibaba Qwen","provider_slug":"qwen","open_weight":true,"tier_band":"demo_cpu","hf_repo":"Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF","params":"0.49B","context_window":32000,"license":"Apache-2.0","commercial_ok":true,"license_restrictions":[],"redistribute_weights":true,"serving":"LIVE on CPU Space via llama.cpp IF weights present (fastest small); else honest label","code_strength":"Smallest real coder; fits cpu-basic","use_case":"Smallest LIVE demo brain","router_role":"DEMO_CPU_SMALL","modalities":["text","code"],"streaming":false,"tier_name":"alloy_demo_cpu","tier":90,"operator_mirrored":false,"ecosystem_mirror":["killinchu"],"honest_stub":true,"notes":"open-weight alloy; LIVE on CPU Space via llama.cpp IF weights present (fastest small); else honest label"}],"roster_source":"szl_llm_registry (open-weight entries only; closed API models filtered out per doctrine)","backend":{"backend":"generative","endpoint":"https://router.huggingface.co/v1","model_serving":"Hugging Face Router — open-weight (server-side only)","primary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","configured":true,"offline":false,"honest":"Real generative inference via the OpenAI-compatible endpoint with the present HF_TOKEN (disclosed, never sent to browser). 2x retry + roster fallback; never a fabricated answer."},"formula_wiring":[{"formula":"C20","name":"Softmax 1/2-Lipschitz (order-stable)","used_for":"model router stability","maturity":"PROVEN (fragment)","plain":"Routing is stable to small changes."},{"formula":"W7-5","name":"PAC-Bayes min<=avg<=max routing envelope","used_for":"router cost/risk bracket","maturity":"CI-green (Mathlib)","plain":"Routing stays between best and worst option."},{"formula":"W5-3 / W7-4","name":"Conformal coverage + rank-count p-value","used_for":"calibrated confidence on a suggestion","maturity":"PROVEN (axiom-free)","plain":"Calibrated confidence — we never report 100% certainty."},{"formula":"P1-P6","name":"Governed agentic loop (Pipeline.lean)","used_for":"the governed run + signed receipt","maturity":"PROVEN (P5 axiom-gated on hash CR)","plain":"Every code action is governed and receipted."},{"formula":"P3","name":"Non-interference (Goguen-Meseguer 1982)","used_for":"untrusted/pasted input cannot flip a verdict","maturity":"PROVEN (axiom-free core)","plain":"Poisoned input can't override safety."},{"formula":"C10/C11/C12","name":"Byzantine / DLS / FLP","used_for":"optional multi-model consensus vote","maturity":"PROVEN (cores)","plain":"Optional agreement vote with a proven safety bound."},{"formula":"F-G5","name":"Bounded-frontier receipt-DAG termination","used_for":"the agent loop provably halts (edge-safe)","maturity":"PROVEN","plain":"The governed run always finishes."}],"honesty":{"models":"OPEN-WEIGHT only (Mistral/Qwen2.5-Coder/DeepSeek-Coder/Codestral/StarCoder2/Llama/Gemma/Phi). No closed weights baked in. No API key required.","no_agi":true,"lambda":"Conjecture 1 (advisory, never the gate)","locked_proven":8,"sandbox":"sandboxed (restricted subprocess) in the HF CPU Space; full seccomp/container isolation on the tower/UDS pod.","offline":"Offline-capable: deterministic local backend + local sandbox + vendored UI; no external API/CDN required at runtime."}}