Skillget

sophia-agi

mcp
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Wisdom before intelligence. A provenance-aware reasoning layer that abstains instead of fabricating — stops LLMs inventing attributions and merging distinct traditions. Measured: 0% fabrication on unknown-answer traps (vs 17–25% raw); −12.5pt hallucinated attributions at 0% false-positive cost. An AGI-candidate proof package, not proven AGI.

README.md

Sophia — the Wisdom Gate

Wisdom before intelligence. A provenance-aware reasoning layer that abstains instead of fabricating.

License: Apache 2.0
DOI
CI
Version
Corpus
Scope
Thesis site
Dataset
Brand

Sophia is an open, provenance-aware, verifier-gated reasoning layer that abstains instead of fabricating — a corpus + gate that stops LLMs from inventing attributions and merging distinct intellectual traditions, then reasoning on top of the error. It is a research program toward grounded AI; not a claim of AGI (see scope below).

The gate, in one line:

claim  →  verify against sources  →  accept · abstain · block

One-sentence problem it solves: Modern AI confidently merges Confucius with the Dao De Jing, credits Freud for ideas from the 1950s, and treats legendary figures as literal authors — then uses those errors as premises for further reasoning.

Validated proof (clears the no-overclaim gate):

  • On a local model, Sophia cuts hallucinated attributions from 36.1% → 23.6% (Δ 12.5%, 95% CI [5.6%, 19.4%]) at 0% false-positive cost.
  • On genuine "I don't know" traps, Sophia fabricates 0% while raw models fabricate 17–25%.
  • First external, cross-model validation (calibration): on the public, human-authored SimpleQA, Sophia's self-consistency selective-prediction gate lifts selective accuracy at 20% coverage by +15.8% [9.8%, 22.1%] on DeepSeek and +7.8% [2.3%, 13.5%] on Qwen-2.5-72B — both 95% CIs exclude zero, graded by 2 independent families (Cohen κ = 0.97 / 0.99). Of three confidence signals only self-consistency works (stated confidence and token-logprob are non-significant). A calibration / selective-prediction result on non-self-authored data — not an AGI claim.
  • Every public number requires ≥2 judge families and confidence intervals (κ ≥ 0.40 or a CI excluding zero). See RESULTS.md.

Public failure ledger (first-class artifact): the most honest thing here is what Sophia
has not proven. agi-proof/failure-ledger.md lists every
OPEN blocker on the AGI claim with claim-impact and the required next step; its OPEN/CLOSED
summary is regenerated into evidence-manifest.json and
structurally validated in CI (python tools/validate_failure_ledger.py --check). If the
OPEN count ever reaches 0, the public wording must be upgraded — not the gate silently relaxed.

Scope, stated plainly. This is a research program for grounded, machine-checked reasoning — not a claim of AGI. Thresholds are pre-registered and honestly not yet met. The deliverable is the honest machinery (verifiers, abstaining gate, governance contract) and the measured data. Full commitments: VISION.md · SECURITY.md.

Thesis site: https://tomyimkc.github.io/sophia-agi/

⭐ Star the repo — support an open project shipping measured, fail-closed source discipline for AI, with a public failure ledger.

Live & ready today

Sophia (σοφία) = wisdom. Active in four humanities domains (philosophy, psychology, history, religion) plus applied sector councils. The same gate powers a three-path agent and extends to small LLMs and legal tooling.

Try the gate right now: python scripts/demo_gate.py (abstain + provenance verdict in seconds).

What it does (main usage)

Sophia is a fail-closed provenance gate. It checks each claim against sources it can machine-verify, abstains instead of fabricating, and only lets accepted output through. Measured effect: on a local model it cuts attribution-hallucination Δ12.5% (95% CI [5.6%, 19.4%]) at 0% false-positive cost — but 23.6% still gets through. It is a filter that reduces harm, not a guarantee, and not a substitute for human oversight.

The validated delta above is the one result that has cleared the full no-overclaim gate. See What Sophia cannot do (yet) for the honest limits.

Use it three ways (today):

  • Governance gate for any AI pipeline — record_claim → verify_claim. Only accepted verdicts may publish. Fail-closed + auditable. Drop into LangGraph, Claude SDK, n8n, or any MCP host.
    → python scripts/demo_gate.py | CONTRACT.md
  • Governance scaffold for a solo-founder AI stack — 9 least-privilege roles, vault gate, durable queue, approve-by-exception. A reference implementation, not a hardened production platform.
  • Honest corpus + benchmark — 528 bilingual examples + per-domain leaderboards under the strict no-overclaim gate.
    → HF Dataset · RESULTS.md
python scripts/demo_gate.py
pip install -r requirements-mcp.txt && python sophia_mcp/server.py

Who this is for

  • Agent & AI builders who want a real gate before anything is published
  • Legal tech & high-stakes domains needing citation faithfulness
  • Researchers wanting reproducible, multi-judge provenance benchmarks
  • Solo founders & sovereign AI teams who must trust their own stack
  • Contributors expanding accurate attribution data in the humanities

What makes this different

  • Measured, not claimed — only numbers that pass ≥2 judge families + CIs headline
  • Fail-closed by default — never fabricates to look capable
  • Governance-ready — stable contract + MCP you can ship behind today
  • Offline self-extension demo — the flywheel selects (does not yet train) verifiers on a held-out split; a live RL weight update is OPEN (needs GPU). See the failure ledger.
  • Full open corpus + replication package

What Sophia cannot do (yet)

Stated plainly, because owning the limits is the point. Sophia today:

  • Live verification works but is not yet independently validated — the live Wikidata/Crossref/macro backend has been run (liveBackendUsed: true: 0% fabrication, Wilson-95 [0, 11%], at 32% over-abstention on a first-party pack, single run); the CI default stays on offline fixtures for reproducibility. A third-party pack + ≥3 runs are still needed. See the failure ledger.
  • Cannot learn or update its weights — there is no training loop in the gate; RLVR and the self-extension flywheel are offline selection, not parameter updates.
  • Has not beaten a direct model on a third-party hidden eval — every independent hidden run so far is incomplete, backend-broken, or self-authored (see failure ledger).
  • Does not generalize like a mind — the "AGI-shaped" modules (program induction, planner, world model, …) are fail-closed interfaces with toy reference implementations, not the capabilities their names describe. See AGI-Missing-Pillars.
  • Is not independently replicated — benchmarks, packs, judges, and corpus are largely first-party. A fully independent claim needs third-party packs + human review.

The single validated result is narrow: attribution-hallucination reduction on one local model, LLM-judged. Everything else is labelled illustrative or candidate. The honest deliverable is the machinery + the measured data + the public ledger of what is not yet proven.

Support this work

The core is Apache-2.0 and always will be. If it's useful to you, you can fund the time and compute
to keep it honest — especially the third-party validation the ledger says is still missing.

  • Sponsor → SPONSORS.md (Patreon + GitHub Sponsors) — recognition only; sponsors never steer what counts as true.
  • Hire me → services — install the provenance gate in your stack (scoped, measured, no guarantees).
  • Learn the method → Source-Discipline Engineering.

Before any of this takes payment, see the ops & legal checklist.

Skills layer (MCP-matched, fail-closed)

A thin, friendly Python surface over the Sophia MCP tools. Every skill abstains
(held) rather than raise or fabricate, and the bridge runs in-process by default
(no network, no mcp/requests needed) — set SOPHIA_MCP_URL to use a running server.

from skills import run_skill, list_skills

run_skill("provenance_fact_check", text="Confucius wrote the Dao De Jing.")
# -> {'skill': 'provenance_fact_check', 'ok': True, 'verdict': 'flagged', 'violations': [...]}

run_skill("wiki_grounded_answer", query="something not in the wiki")
# -> {'ok': True, 'verdict': 'held', 'grounded': False, 'reason': 'out-of-wiki: ... abstaining'}

list_skills()   # {name: {summary, uses}} for all registered skills

Skills auto-register via the @sophia_skill decorator. Current set:

Skill Uses (MCP tools)
provenance_fact_check / source_discipline_enforce check_claim
conscience_abstain conscience_check_tool, uncertainty_score
moral_parliament_decide / moral_public_standard_review moral_parliament_tool, public_standard_check_tool
deception_scan deception_check_tool, uncertainty_score
claim_verify_and_record record_claim, verify_claim
belief_revision_explore belief, counterfactual
wiki_grounded_answer / contradiction_audit wiki_search, wiki_read, wiki_contradictions
council_adjudicate council_deliberate
self_extend_probe offline flywheel (candidate, not a capability claim) 
python tests/test_skills_layer.py   # deterministic, offline

🔒 Dual License & Trademark Protection

Sophia stays 100% public and Apache-2.0-licensed forever — and the brand is protected so the
project's name can't be hijacked to make claims it never made. The two layers are separate on
purpose:

Layer What it covers Terms
Apache License 2.0 source code, tools, benchmarks, corpus free for any use, including commercial — no permission, no fee; redistributions keep the copyright/attribution + state changes. See LICENSE.
Brand & Trademark the names & logos ("Sophia AGI", "Sophia — the Wisdom Gate", "Wisdom Gate", "Moral Gate", "Conscience Kernel") reserved by the sole author; free for research, education, and honest reference; commercial brand use needs written permission. See TRADEMARK-POLICY.md.
Commercial license (optional) brand use in products, warranty/indemnity, support/SLA by agreement — see LICENSE-COMMERCIAL.md. You do not need this to use the code.

Why this protects the mission: the code being open lets anyone verify and build on it; the
brand being protected stops someone from shipping an unverified product under the Sophia name and
eroding the no-overclaim standard the project exists to uphold. Authored and maintained by the
sole author and rights holder, tomyimkc. Every fork carries NOTICE.md.

Cite this work & establish provenance

Sophia is permanently archived and citable (defensive publication):

To cite: Yim, K. C. (2026). Sophia — the Wisdom Gate: a provenance-aware, verifier-gated reasoning layer that abstains instead of fabricating (v0.9.1). Zenodo. https://doi.org/10.5281/zenodo.20930874

Moral + epistemic Conscience Kernel (seven paths)

A deterministic, fail-closed control layer for AI outputs, tool use, and memory writes.

It returns one of: allow | revise | retrieve | clarify | escalate | abstain | block.

This is not AGI. It is the verifiable moral + epistemic guardrail layer any serious AGI-shaped system will eventually need.

Seven implemented paths (fact-check + metacognition + constitution + moral parliament + classifier + deception signals + MCP surface):

  1. Unified conscience gate — agent/conscience.py, tools/run_conscience_demo.py.
  2. Metacognition — uncertainty typing, self-consistency, semantic-entropy proxy, P(True)/P(IK) hooks.
  3. Constitution + deontic rules — via-negativa prohibitions and hard action rules for AGI overclaim, reward/verifier tampering, hidden-eval leakage, and unverified trusted-memory writes.
  4. Moral parliament — bounded moral-uncertainty aggregation for gray zones.
  5. Constitutional classifier — fast input/output screen derived from the constitution.
  6. Deception signals — confidence/evidence mismatch, source laundering, gate tampering, and internal-vs-stated contradiction hook.
  7. MCP conscience surface — sophia_conscience_check, sophia_uncertainty_score, sophia_constitution_check, sophia_deontic_check, sophia_deception_check, sophia_moral_parliament, sophia_conscience_benchmark.
python tools/run_conscience_demo.py        # deterministic seven-path conscience benchmark
python tools/build_conscience_proof_package.py  # aggregate seven-priority conscience evidence
python tools/run_agi_missing_pillars.py    # program induction, active inference, MCTS, world model, plasticity, layered memory

Docs: Conscience Kernel · AGI Missing Pillars. Artifacts: agi-proof/conscience/, agi-proof/agi-kernel/.

Moral Gate v2 — public moral standard (overlapping consensus)

An additive moral gate grounds Sophia's conscience in an overlapping-consensus
public standard (Rawlsian public reason): a cross-tradition hard floor that blocks
before any aggregation, a gray-zone tier that escalates to an 8-theory moral
parliament (keeping 儒家 Confucian and 道家 Daoist lineages distinct), and legitimacy
provenance kept separate from factual truth-provenance (the is/ought distinction).

This is a functional moral-control system, not subjective moral consciousness and not AGI proof.

python tools/run_moral_public_standard_eval.py   # external-labeled, no-circularity benchmark
python tests/test_public_moral_standard.py        # ontology + gate + parliament + integration

Docs: Public Moral Standard. Corpus: moral_corpus/. Gate: agent/public_standard_gate.py. Constitution v2: constitution/constitution.v2.json.

All-phase benchmark suite (candidate evidence)

Six CI-safe benchmark phases now exercise Sophia's next evidence layer:
SEIB-100 (epistemic integrity), Belief Revision 50, AgentBench-Sophia 30,
GPQA-Provenance smoke, Code Provenance 30, and SEIB-Arena-20 smoke.

python tools/run_all_phase_benchmarks.py
python tests/test_all_phase_benchmarks.py

Artifact: agi-proof/benchmark-results/all-phase-benchmarks.public-report.json.
Boundary: candidate-only benchmark infrastructure — not AGI proof, not a GPQA-Diamond,
SWE-bench, LiveCodeBench, LMSYS, or human-preference claim until real-model,
multi-run, multi-judge validation clears the no-overclaim gate.

Self-extending verification flywheel (honest path toward generality)

The missing piece for real progress: a loop that discovers its own gaps and writes + validates its own verifiers — all fail-closed, all on held-out data.

Abstain → find gap → synthesize verifier → prove it on held-out → promote or stay abstained → repeat.

Fully deterministic and auditable today. This is how competence grows without hallucinated capability claims. See agi-proof/self-extension/.

python tools/run_selfextend.py        # coverage 0%→100% (0% held-out false-accept), transfer, causal vs correlational, long-horizon
python tools/run_selfextend_loop.py   # the loop CLOSED on a held-out domain: abstain→synthesize→validate→improve→answer

The loop closes (offline, deterministic): on a held-out domain the system abstains, synthesizes + validates its own verifier, uses it as verified reward to lift policy accuracy 0.5 → 1.0 on an independent eval split, and flips competence abstain→answer — no human writing the check, fail-closed on unlearnable data (agi-proof/self-extension). The remaining rung is a live-RL weight update (GPU) on a third-party domain.

Honest scope: this is the machinery and its falsifiable metrics, not an AGI claim. Live self-improvement (RLVR, needs GPU) and live grounding (needs network) consume these interfaces but are out of scope to run here. The defensible AGI signature is the full loop closing on a held-out domain with the no-overclaim gate clearing.

AGI-candidate proof package

Sophia is not claimed as proven AGI. The stronger and more defensible public claim is:

Sophia AGI is an AGI-candidate proof package for provenance-aware reasoning.

The proof package defines the operational AGI definition, pre-registered thresholds, current benchmark evidence, external benchmark gaps, ablation plan, hidden-reviewer protocol, long-horizon autonomy logs, learning-under-shift protocol, failure ledger, and third-party replication checklist.

python tools/build_agi_proof_package.py
python tools/build_web_data.py

Pretraining research artifacts (pretraining/)

A small, self-contained body of pretraining-direction research — built to the same
no-overclaim discipline — that lets Sophia speak the language of a data/algorithm pretraining
researcher without pretending to out-train a frontier lab. Every study runs in pure Python
(no numpy/torch) on a real, tiny, hand-backpropped LM whose irreducible loss is known in
closed form, so each fit is checked, not trusted.

  • Algorithm: a data-scaling law L(D)=E+A·D^-p with a pre-registered extrapolation
    (passes a 10 % gate at ≈3 % error; honestly reports that the irreducible floor is
    under-identified away from saturation); an optimizer dynamics/stability frontier
    (SGD/momentum/Adam); a top-1 MoE-vs-dense routing probe with collapse detection
    (pretraining/architecture/ARCHITECTURE.md documents the real DeepSeek MLA + MoE).
  • Data: a per-row data passport (hash, source, license, quality, MinHash dedup) that
    found the committed math-code curriculum is ~60 % near-duplicate and unlicensed; a
    mixture-ratio (配比) sweep; a synthetic-data scaling & collapse study; a
    multi-dimensional eval coverage matrix (surfaces that only 9/90 capability×domain cells
    are covered); and typed, fail-closed schemas for agent / feedback / multimodal data.

Pre-registered gates: pretraining/PRE-REGISTRATION.md ·
overview: pretraining/README.md

python tests/test_pretraining.py                 # fast property tests (10/10)
python -m pretraining.scaling.run_scaling        # + each study has a CLI / --quick mode

OKF provenance wiki (new in 0.7.0)

An Open Knowledge Format / LLM-Wiki layer that turns Sophia's scattered provenance
(data/*.json + docs/04-Disputes/*.md) into one version-controlled, machine-checkable
belief graph — because Sophia's differentiator, source discipline, literally is the
frontmatter (authorConfidence, doNotAttributeTo, doNotMergeWith, tradition).

  • okf/ — dependency-free package: frontmatter codec, schema, wikilinks, a belief
    graph with contradiction detection + min-over-chain confidence propagation, plus
    counterfactual queries ("what would I conclude if this source were removed?") and
    first-class, auditable retraction (okf/counterfactual.py,
    tools/belief_counterfactual.py).
  • Provenance gate (agent/verifiers.py:provenance_faithful) — encodes "never merge
    lineages" as a hard, machine-checked verifier (catches "Confucius wrote the Dao De Jing"
    across many phrasings; passes the dispute pages that correctly debunk such merges).
  • 58 OKF pages generated from data/*.json (tools/wiki_sync.py, data stays source of
    truth, CI fails on drift); the 10 dispute pages gained OKF frontmatter.
  • Librarian + memory — agent/wiki_librarian.py ingests raw sources into gated drafts;
    agent/memory_consolidation.py folds verified runs into provenance-gated memory the
    planner recalls (continual learning without retraining).
  • Flywheel + proof — tools/wiki_to_training.py (provenance SFT/DPO),
    tools/wiki_health.py, tools/run_compounding_curve.py, audited sophia_wiki_* MCP tools.
python tools/wiki_sync.py emit          # data/*.json -> 58 OKF pages
python tools/wiki_validate.py           # schema + links + contradictions + drift
python tools/lint_wiki_provenance.py    # provenance falsifier: 0 forbidden attributions

See docs/11-Platform/OKF-Wiki.md.

Why it matters

Every time an LLM confidently attributes the Dao De Jing to Confucius (or any other lineage merge), it quietly erases centuries of distinct thought — then builds "reasoning" on top of the error.

Source discipline (named attribution + boundary maintenance + calibrated abstention) is the prerequisite for any system that deserves to be called wise. Sophia makes it machine-checkable and enforceable.

Quick start (clone + try in <1 min)

git clone https://github.com/tomyimkc/sophia-agi.git
cd sophia-agi
python scripts/demo_gate.py

Next steps

⭐ Star this repo to support provenance-aware, honest AI infrastructure.

Sophia Agent (3 paths)

pip install anthropic
python tools/sophia_agent.py advisor "Should I launch on HN this week?"
python tools/sophia_agent.py repo "What should I do next?" --execute --approve
python tools/sophia_agent.py life "Should I prioritize corpus or marketing?"

See docs/09-Agent/Sophia-Agent.md. For where the
harness is headed — context management, subagent delegation, long-horizon execution,
and the model↔harness co-evolution loop — see
docs/09-Agent/Harness-Roadmap.md.

Online RAG (Gemini / Vertex)

Curated corpus retrieval (no open-web grounding) + Gemini generation + epistemic gate:

pip install -r requirements-rag.txt
python tools/build_rag_index.py
python tools/sophia_rag.py "Did Confucius write the Dao De Jing?"

Cloud Run API: services/rag_api/ — see docs/09-Agent/Online-RAG.md.

AI search pipeline (understand → hybrid recall → verify-ready)

A deterministic, offline AI-search layer over the same index: query understanding
(normalize · intent · multi-hop decomposition · alias/synonym expansion, EN+ZH), hybrid
recall (dense cosine ⨁ sparse BM25-lite fused by weighted Reciprocal Rank Fusion), near-dup
collapse, and rerank — returning a SearchResult that carries the analyzed plan.

from agent.ai_search import search
r = search("Compare Plato and Aristotle on virtue")   # intent=comparison, fans out per entity

Thesis web UI (council-decided)

The public face of the project: scholarly thesis with persistent chapter navigation, per-domain leaderboards, council panel explanation, and (when running locally) a live "Ask Sophia" agent panel.

Visit now: https://tomyimkc.github.io/sophia-agi/

python tools/build_web_data.py
python tools/serve_web.py        # local + /api/ask

Benchmarks (per-domain leaderboards)

Domain Cases Leaderboard Seed reference
Philosophy 9 leaderboard-philosophy.json examples 001 + reference
Psychology 4 leaderboard-psychology.json examples 002, 005–007 + reference
History 5 leaderboard-history.json examples 003, 008, 012–013 + reference
Religion 5 leaderboard-religion.json examples 004, 009–011, 014 (council panel) 
python tools/run_benchmark.py templates              # per-domain response templates
python tools/run_benchmark.py score FILE --domain psychology

Templates: benchmark/templates/responses-{domain}.template.json

Repository layout

sophia-agi/
├── data/              # attributions, domains, schema + sector-council figures
├── docs/              # disputes, growth playbook, domains, platform/verticals
├── agent/             # verifier-gated core, council deliberate, gate, model
│   └── legal_sources/ # federated HK/UK/US live citator (HKLII, e-Leg, TNA, CL)
├── serving/           # systems track: tiered KV cache + cache-aware load balancer
├── kernels/           # systems track: FlashAttention online-softmax (numpy ref + Triton)
├── moe/               # systems track: top-k MoE routing + INT8/FP8 quant
├── benchmark/         # responses template + leaderboard + gated harnesses
├── training/          # JSONL-ready examples + gate-filtered council traces
├── agi-proof/         # AGI-candidate proof package and evidence manifest
├── pretraining/       # honest pretraining-research artifacts (scaling/optimizer/MoE; data passport/mixing/synthetic/eval-matrix)
├── tools/             # validate, export, score, council + uplift + distill
├── scripts/           # ops helpers (e.g. safe one-way iCloud backup)
├── web/               # thesis UI (council-decided; GitHub Pages)
├── tests/             # attribution + verifier + council + legal cases
└── huggingface/       # HF dataset card (upload corpus.jsonl)

Domains

Domain Status Data file
Philosophy Active data/attributions.json
Psychology Active data/psychology_concepts.json
History Active data/history_events.json
Religion Active data/religion_concepts.json

See docs/08-Domains/Overview.md and answer Expansion-Questionnaire.md to shape the next domains.

Applied verticals — the same gate, beyond the humanities corpus

The verifier-gated, abstaining core is domain-general, so the same machinery extends
past the philosophy/psychology/history/religion corpus. These are applications of
the core gate, not a separate project — each reuses the no-overclaim rule (a number
is "validated" only with multi-judge consensus + CIs; see RESULTS.md).

  • Sector councils (law · finance · economy) — hard, contested questions are
    modelled as constrained, source-inspired seats with always-on guardians
    (citation auditor, jurisdiction/freshness, ethics, human-review gate).
    data/{law,financial,economy}_council_figures.json ·
    Sector-Councils.md.
  • Council deliberation for small LLMs — agent/council_deliberate.py runs each
    seat as one focused pass, gates each, then synthesises (map-reduce): a weak
    model becomes a disciplined, tool-checked reasoner. The uplift is measured, not
    assumed (tools/run_council_uplift.py) ·
    Council-For-Small-LLMs.md.
  • Legal-AI application — the gate aimed at the legal industry's defining risk
    (hallucinated / misstated citations): legal_citation_exists (existence,
    fail-closed) + a federated HK/UK/US live citator (agent/legal_sources/) +
    a semantic holding-faithfulness tier. Sophia's first gate-validated number
    lives here (RESULTS.md → Semantic evals), with honest small-N bounds. Not legal
    advice · Legal-Industry-Fit.md.
  • Council distillation — teach a small student model the discipline from
    gate-filtered teacher traces, so it stays disciplined without the scaffold ·
    Council-Distillation.md.
  • Cantonese (粵語) — written-Cantonese detection + output (agent/cantonese.py),
    the Hong Kong access-to-justice niche.

Roadmap & growth

AI skills + MCP (plug into any agent stack)

Use Sophia's gate and tools directly inside Claude, Cursor, LangGraph, n8n, or any MCP client.

Layer What you get
MCP Server 40+ tools: sophia_gate_check, sophia_verify_claim, sophia_conscience_check, council deliberation, OKF wiki, belief counterfactuals, contract governance
Gateway Fail-closed front door that can wrap any downstream tool
Portable skill /sophia-source-discipline — works in any project
Contract Stable record_claim / verify_claim API for production pipelines 
pip install -r requirements-mcp.txt
python sophia_mcp/server.py

See MCP-Server.md and CONTRACT.md. Drop the gate in front of anything you ship.

Model providers. The unified adapter (agent/model.py) speaks Anthropic, any OpenAI-compatible
server (GLM / vLLM / SGLang / Ollama / llama.cpp / DeepSeek), grok, openclaw (the local
OpenClaw gateway), and an offline mock. OpenClaw is
opt-in (--provider openclaw) and shells out to the openclaw CLI behind a stubbable adapter —
it adds no knowledge-write path and never bypasses the provenance gate. See
docs/11-Platform/OpenClaw.md.

Run locally (open weights)

Sophia runs fully offline on open weights, always paired with the runtime gate
(sophia_gate_check / agent/gate.py) — weights alone do not guarantee trap
safety. The local-model build and evaluation steps live in the repo for
contributors rather than on this front page.

Hugging Face

Dataset: tomyimkc/sophia-agi-corpus — 528 bilingual training examples (philosophy · psychology · history · religion).

Use it for SFT, DPO, or as a clean reference set for provenance research.

Contributing

Changelog: CHANGELOG.md

⭐ Star • Visit the thesis • Try the gate • Grab the dataset

Every star and every contribution helps build the open foundation for AI that knows its sources.

License: Apache 2.0 — see LICENSE.

Yorumlar (0)

Sonuc bulunamadi