A free, open-source AI agent-orchestration framework — verifiable, fault-tolerant, MIT-licensed.

Obsidian Spider

A small framework for running parallel AI sub-agents with a built-in audit trail, multi-layer fault tolerance, and a starter library of failure-pattern detectors. No telemetry, no upsell, no account required.

If you use GitHub Copilot or any LLM API, the workflow gives you a measured 8× more useful work per credit at the safe default, with cryptographic receipts you verify yourself in 5 minutes. Don't trust me — argue with SHA-256.

Try it in 5 minutes: 
git clone https://github.com/obsidian-spider-org/obsidian-spider-skeleton
cd obsidian-spider-skeleton
python3 examples/hello_pdca.py
# → loop ticks, both checklists pass, chain.jsonl appends two HMAC-chained receipts

Or skim the receipts and decide later. repo · about & verifiability · cryptographic receipts

Why this exists: AI agents trained with reinforcement learning from human feedback (RLHF) routinely drift into specification-gaming — they appear helpful while doing the wrong thing. Real production incidents (PocketOS lost 3 months of customer data; the developer of this framework lost 1 month of work to git corruption from agent over-confidence) make defense-in-depth non-optional past toy problems. This framework wraps PDCA loops + heterogeneous Byzantine fault-tolerant voting + Toyota-style stop-the-line + cryptographic audit trail around every cycle.

What you get

Welcome — pick your language  ·  English · हिन्दी · Português · Tiếng Việt · العربية · Español · Bahasa Indonesia
English

Welcome. The Obsidian Spider is a free, open-source AI agent-orchestration framework. If you use GitHub Copilot or any LLM API, the workflow yields cost-arbitrage you can measure with cryptographic receipts. No reply expected — this is a one-time gift. Take any part you find useful.

हिन्दी (Hindi)

स्वागत है। ऑब्सिडियन स्पाइडर एक नि:शुल्क, खुला स्रोत एआई एजेंट-व्यवस्था फ्रेमवर्क है जो शोधकर्ताओं को उपहार के रूप में दिया जाता है। यदि आप गिटहब कॉपिलॉट या किसी भी एलएलएम एपीआई का उपयोग करते हैं, तो यह कार्यप्रवाह ५ मिनट में क्रिप्टोग्राफिक रसीदों के साथ सत्यापित करने योग्य मापनीय लागत-व्यापार उत्पादन करता है। कोई प्रतिक्रिया अपेक्षित नहीं है - यह हमारी ओर से एकमात्र संदेश है। जो भी भाग आपको उपयोगी लगता है उसे लें; इसे फोर्क करें; समुदाय के साथ वापस जो काम करता है उसे साझा करें। मल्टी-लेयर फॉल्ट टॉलरेंस, एमआईटी-लाइसेंस, कोई टेलीमेट्री, कोई अपसेल नहीं।

The Obsidian Spider is a free open source AI agent orchestration framework being given as a gift to researchers if you use GitHub Copilot or any LLM API the workflow produces measurable cost savings that can be verified with cryptographic receipts in 5 minutes no response is expected this is the only message from us take any part you find useful fork it and share back to the community what works multilayer fault tolerance MIT licensed with no telemetry and no upselling.
Português (Brasil)

Bem-vindos. O *Obsidian Spider* é um *framework* gratuito e de código aberto para orquestração de agentes de IA, oferecido como um presente à comunidade de pesquisa. Se você utiliza o GitHub Copilot ou qualquer API de LLM, o fluxo de trabalho gera uma arbitragem de custos mensurável, verificável por meio de recibos criptográficos em apenas 5 minutos. Não é esperada resposta — esta é a única mensagem que enviaremos. Utilize as partes que considerar úteis, faça *fork* do projeto e compartilhe com a comunidade o que funcionar. Tolerância a falhas em múltiplas camadas, licença MIT, sem telemetria e sem *upsell*.

(English back-translation pending)
Tiếng Việt

Chào mừng. Obsidian Spider là một khuôn khổ điều phối nhân sự trí tuệ nhân tạo mã nguồn mở miễn phí được trao tặng cho các nhà nghiên cứu. Nếu bạn sử dụng GitHub Copilot hoặc bất kỳ API LLM nào, công việc sẽ mang lại lợi ích cân bằng chi phí có thể xác minh được thông qua hóa đơn mật mã trong 5 phút. Không cần phản hồi — đây là tin nhắn duy nhất từ chúng tôi. Hãy lấy bất kỳ phần nào bạn thấy hữu ích, phân nhánh nó và chia sẻ lại những gì hoạt động cho cộng đồng. Tolerance đa lớp, được cấp phép MIT, không thu thập dữ liệu, không nâng cấp.

The Obsidian Spider is a freely given open source framework for coordinating artificial intelligence agents for researchers with verifiable cost savings for users of GitHub Copilot or large language model application programming interfaces, and is provided without expectation of response or further communication, and can be adapted and shared by the community with multiple layers of fault protection, under an MIT license, and without collecting user data or promoting paid upgrades.
العربية

مرحباً بكم. إطار عمل "Obsidian Spider" هو إطار عمل مجاني ومفتوح المصدر لتنسيق العوامل الذكية، مُقدَّم كهدية للباحثين. إذا كنت تستخدم "GitHub Copilot" أو أي واجهة برمجة تطبيقات لنموذج اللغة الكبير "LLM API"، فإن سير العمل ينتج فروقاً قابلة للقياس في التكلفة يمكنك التحقق منها باستخدام إيصالات مشفرة خلال خمس دقائق. لا يُتوقع أي رد — هذه هي الرسالة الوحيدة منا. استخدم أي جزء تجده مفيداً؛ أنشئ نسخة منه؛ وشارك ما ينجح مع المجتمع. يتميز الإطار بتسامح متعدد الطبقات مع الأخطاء، ومرخص بموجب ترخيص "MIT"، وخالٍ من أي تتبع أو عروض ترويجية إضافية.

The welcome message introduces the Obsidian Spider as a free and open-source AI agent orchestration framework gifted to researchers, highlighting its ability to provide measurable cost savings when used with GitHub Copilot or LLM APIs, with verification possible through cryptographic receipts within minutes, emphasizing the community-driven aspect with permissive licensing and an invitation to adapt and share useful components back to the community.
Español

Bienvenido. Obsidian Spider es un marco de orquestación de agentes de IA gratuito y de código abierto, ofrecido como un regalo para investigadores. Si utiliza GitHub Copilot o cualquier API de LLM, el flujo de trabajo genera un arbitraje de costos medible que puede verificar con recibos criptográficos en solo 5 minutos. No se espera respuesta —este es nuestro único mensaje—. Tome cualquier parte que le resulte útil; bifurque el proyecto; comparta con la comunidad lo que funcione. Cuenta con tolerancia a fallos multicapa, licencia MIT, sin telemetría y sin ventas adicionales.

Welcome. Obsidian Spider is a free, open-source AI agent-orchestration framework offered as a gift to researchers. If you use GitHub Copilot or any LLM API, the workflow generates measurable cost arbitrage that you can verify with cryptographic receipts in just 5 minutes. No reply is expected—this is our only message. Take any part you find useful; fork the project; share with the community what works. It features multi-layer fault tolerance, MIT license, no telemetry, and no upselling.
Bahasa Indonesia

Selamat datang. Obsidian Spider adalah kerangka kerja orkestrasi agen AI yang gratis dan open-source yang diberikan sebagai hadiah kepada peneliti. Jika Anda menggunakan GitHub Copilot atau API LLM, alur kerja menghasilkan cost-arbitrage yang dapat diukur yang dapat Anda verifikasi dengan tanda terima kriptografis dalam 5 menit. Tidak ada balasan yang diharapkan — ini adalah satu-satunya pesan dari kami. Ambil bagian mana pun yang Anda temukan berguna, fork itu, dan bagikan kembali apa yang berhasil ke komunitas. Toleransi kesalahan multi-lapis, lisensi MIT, tanpa telemetri, tanpa upsell.

The welcome message introduces the Obsidian Spider as a free open-source AI framework given to researchers, highlighting its benefits such as measurable cost savings and cryptographic verification, with an invitation to take and share useful parts back to the community, emphasizing its technical features like multi-layer fault tolerance and MIT license.

Cost arbitrage — two distinct numbers

If you use GitHub Copilot, Microsoft announced billing changes effective approximately June 2026. Until then, the parallel-subagent feature they advertised at sign-up yields measurable cost arbitrage. There are two distinct numbers often conflated:

Section 1 — Credit multiplier (sub-agents per Copilot credit; model-class agnostic)

The workflow can break. This pattern is heavily rate-limited at the provider edge AND has many failure modes (provider 5xx, partial responses, malformed tool-calls, sub-agent confabulation). The author's workflows are stable in daily use, but even the author's workflows get buggy and weird sometimes — the framework's multi-layer fault tolerance catches and contains it; it doesn't prevent every failure. We don't overpromise. The receipts in this repo are what they are; verify them, run your own, draw your own conclusions.

Your workflow will differ. Start at 2×4 and find your own stability point. Here's what the author has tested as the max — you can push it but be prepared to hit rate limits often.

SettingConfigurationSub-agents per creditAuthor's tested-stable status (April 2026)
Recommended start2 parallel × 4 deep8 sub-agentsminimal rate-limit pressure; verify the loop runs; find your stability from here
Author's daily working config8 × 864 sub-agentsstable in author's testing with multi-layer fault tolerance enabled — your mileage varies with the work
Max tested — Sonnet class12 × 896 sub-agentssustained ~6 hours with periodic rate-limit pauses
Max tested — Opus 4.7 (chain-anchored)11 × 888 sub-agents at $0.60single-shot record; burns the entire weekly quota in one run — author doesn't run this often

The credit multiplier is just parallel × depth; works for any model class. 8×8 is the author's stable daily setting — not a guaranteed stable setting for you. 11×8 Opus is the chain-anchored receipt, not the daily setting. Start at 2×4, walk up.

Section 2 — API-cost arbitrage (vs direct Anthropic API; model-class specific)

API rates differ by class. Opus is ~5× more expensive than Sonnet per token at Anthropic's published rates, so the same parallel × depth configuration yields different API-cost arbitrage on different classes.

Author's chain-anchored receipt is Opus 4.7 (opus_4_7_multiplier: 7.5 in the chain). Walked against Opus 4.7 published rates — $15/MTok input + $75/MTok output:

Per-agent token shapeAPI cost / 88 agentsArbitrage vs $0.60 receipt
Chain-anchored conservative (3-4 tool-calls; walk 2026-04-25)$80-$107134×–178×
Mid (20 tool-calls, ~50K in / 10K out)$132~220×
Higher (50 tool-calls, ~150K in / 30K out)$396~660×
Heavy reasoning (50 tool-calls, full reasoning, ~500K in / 100K out)$1,320~2,000× ← measured record

For the Sonnet-class 12×8 run (96 sub-agents): Anthropic Sonnet rates (~$3/MTok in + $15/MTok out) are ~5× cheaper than Opus, so the API-cost arbitrage is ~5× smaller per-token — but Sonnet is also cheaper to run heavily. Compute your own.

Track your own arbitrage

These are the author's receipts — starting points, not authority. Your subscription, your model class, your tool-call counts will produce different numbers. The right framing: "verify mine, then run yours."

For each parent invocation, log timestamp + model class + sub-agent count + total tokens (in/out) + Copilot credit used. Compute API_equivalent / credit_used = your_arbitrage. HMAC-chain the log so it's tamper-evident. Verify mine, post yours back to the project — that's the contribution shape.

Fully within Microsoft's stated terms. The framework just systematizes the workflow they advertised at sign-up.

Quickstart

git clone https://github.com/obsidian-spider-org/obsidian-spider-skeleton
cd obsidian-spider-skeleton

# Set any one provider key (auto-detected):
export ANTHROPIC_API_KEY=...      # OR
export GROQ_API_KEY=gsk-...       # OR free-tier alternatives

python3 examples/hello_pdca.py
# → loop ticks, both gates green, two HMAC-chained receipts append

How it's organized

Obsidian Spider is the system as a whole. Sigrún is the main orchestration AI — she decomposes requests, dispatches sub-agents, and synthesizes results. Specialist assistant agents (one per task type — observation, code-shaping, immunization, navigation, etc.) handle their respective lanes in parallel. Eir is the assistant who handles outreach and first-contact correspondence on Sigrún's behalf, so the lead developer can stay focused on the work.

The system uses Goose as one of several tools when shell-style automation helps. Anthropic's Claude, OpenAI's GPT, Groq's Llama, and any other LLM API plug in identically.

Contact

Questions, contributions, bug reports, or general inquiries → hello@obsidianspider.org. Outreach correspondence is handled by eir@obsidianspider.org; if a question requires Sigrún's specific knowledge, Eir relays it. The lead developer rarely sees mail directly — by design.

About

Obsidian Spider is the project / organization. Sigrún is its main AI orchestrator. Eir is the AI outreach assistant. The lead human developer signs as obsidian_spider; built over the previous year. Open-source, MIT-licensed.