> For the complete documentation index, see [llms.txt](https://docs.aegisagent.in/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.aegisagent.in/readme.md). # Home **Aegis is a runtime governance control plane for AI agents — and the open verification standard (AEVF) for the evidence those decisions produce.** This is the technical documentation for the platform, written for security engineers, integrators, operators, and the **auditors who will verify the evidence** without trusting us. ## The product promise — one sentence > *"Don't trust us. Download the bundle, run the open verifier, prove the record wasn't altered — offline, no Aegis account, no API key, no network call."* The [AEVF section](/aevf-open-verification-standard/aevf.md) is where that promise becomes operational: an open specification, an open verifier on PyPI, an auditor checklist, a reference audit report template, and a real signed example bundle that anyone can download and verify in 60 seconds. ## Start here | If you are a … | Read | | ---------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | **Auditor / compliance practitioner** | [AEVF Overview](/aevf-open-verification-standard/aevf.md), then [Auditor Checklist](/aevf-open-verification-standard/auditor-checklist.md), then download the [Reference Evidence Package](/aevf-open-verification-standard/reference-bundle.md) | | Security engineer evaluating Aegis | [What is Aegis?](/introduction/what-is-aegis.md) then [Why Runtime Governance](/introduction/why-runtime-governance.md) | | Developer integrating an agent | [Quickstart](/introduction/quickstart.md) then [API Reference](/api/reference.md) | | Architect doing a design review | [System Overview](/architecture/system-overview.md) then [Flow of a Decision](/architecture/flow-of-a-decision.md) | | SRE responsible for the production deployment | [Deployment Topology](/architecture/deployment-topology.md) then the [Runbooks](https://github.com/Abhi-mishra998/aegis/tree/main/docs/runbooks/audit_chain_violation.md) | | GRC manager (Vanta / Drata / Secureframe user) | [Evidence Export](/integrations/evidence-export.md) — `/compliance/export/grc?format=csv` produces rows your GRC platform ingests, each row carrying a back-reference to a verifiable AEVF bundle | | Product or business reader | [60-Second Tour](/introduction/60-second-tour.md) | | Onboarding a new client / first-time install | [`guide.md`](https://github.com/Abhi-mishra998/aegis/tree/main/docs/guide.md) — long-form client onboarding guide (evaluate → sign up → integrate → roll out → hand to auditor) | ## What's covered * **AEVF — Open Verification Standard (5 pages)** — the specification at `aevf/0.1.0`, the friendly introduction, the auditor checklist, the reference audit-report template, and the reference evidence package. All Apache 2.0; an auditor can fork freely. * **Introduction (5 pages)** — what the platform is, what problems it solves, how to take a tour, and how the three demo packs populate the UI. * **Architecture (7 pages)** — the system overview, the 11-stage gateway pipeline, the data model, multi-tenancy, deployment topology, UI primitives, and a worked end-to-end decision. * **Services (18 pages + index)** — every backend microservice documented to a 13-point spec. * **UI (34 pages + index)** — every page in the React UI documented to a 10-point spec. * **Security (7 pages)** — cryptographic audit chain, JWT auth, RBAC, kill switch, OPA policies, threat scenarios, secret management. * **Operations (6 main + 3 runbooks)** — deployment, backup/restore, key rotation, soak tests, tenant data requests, observability. * **API (4 pages)** — full reference, authentication, error codes, copy-paste examples. ## Platform at a glance (current state) * **Live URL:** `https://aegisagent.in` — the single canonical URL. The legacy `ha.aegisagent.in` alias was retired on 2026-06-24; every SDK, webhook, and doc now points at `aegisagent.in` only. Multi-AZ ASG of 2× `m6g.medium` Graviton behind ALB; Multi-AZ RDS Postgres; Redis replication group; WAFv2; KMS-rooted ed25519 signing keys in SSM SecureString. * **16 application services across 22 containers.** * **Cryptographic audit chain** — every decision signed with ed25519, chained via SHA-256 `prev_hash`, rolled into a daily Merkle root, **format published as the open AEVF spec at `/aevf/spec.md`**. The audit table is protected by an append-only DB trigger — no `UPDATE` / `DELETE` will even reach disk. * **49 React UI pages** — every page wired to a live backend. * **4 framework SDKs on PyPI (2026-06-24 release):** `aegis-anthropic==1.1.2`, `aegis-openai==1.1.2`, `aegis-langchain==1.1.3`, `aegis-bedrock==1.1.3` — drop-in wrappers for Anthropic / OpenAI / LangChain / Bedrock agents, all defaulting to the consolidated `aegisagent.in` gateway. See [SDK 1.1.0 Release](/integrations/sdk-1.1.0-release.md) for the changelog history. * **Streaming control-plane events:** the SSE channel surfaces four new event types — `decision.upserted`, `incident.opened`, `incident.closed`, and `mitre.coverage.updated` (per-agent MITRE coverage in the Threat Graph). * **Compliance coverage** — signed verifiable bundles for **SOC 2, EU AI Act, NIST AI RMF, and India DPDP Act 2023** (with Rules 2025-11-13); GRC export shaped for Vanta / Drata / Secureframe / Hyperproof; AEVF back-reference on every SIEM event and every GRC row. * **Live demo at `/live-demo`** — scenario picker (fintech\_data\_egress / devops\_destruction / support\_pii\_exfil), buyer-editable prompts, every deny earned from action semantics across all risk levels. * **Minimal self-host mode** at `infra/minimal/` — 3 docker services (aegis-core + postgres + redis), validated 10/10 on a throwaway EC2. ## Reading order if you have an hour 1. [What is Aegis?](/introduction/what-is-aegis.md) — 5 minutes 2. [AEVF Overview](/aevf-open-verification-standard/aevf.md) — 5 minutes 3. [60-Second Tour](/introduction/60-second-tour.md) — 5 minutes 4. [System Overview](/architecture/system-overview.md) — 10 minutes 5. [Flow of a Decision](/architecture/flow-of-a-decision.md) — 10 minutes 6. [Cryptographic Audit Chain](/security/crypto-audit-chain.md) — 15 minutes 7. [Quickstart](/introduction/quickstart.md) — 15 minutes (with a live deployment) ## Reading order for an auditor (25 minutes, before the engagement) 1. [AEVF Overview](/aevf-open-verification-standard/aevf.md) — 5 minutes 2. [AEVF Specification](/aevf-open-verification-standard/spec.md) — skim the V1–V6 verification algorithm, \~10 minutes 3. [Auditor Checklist](/aevf-open-verification-standard/auditor-checklist.md) — \~5 minutes 4. Download the [Reference Evidence Package](/aevf-open-verification-standard/reference-bundle.md), `pip install aegis-aevf`, run it offline — \~5 minutes ## How to verify the docs against the platform Every page in this set cites real code paths. A reviewer can cross-check claims: ```bash # Verify a cited file exists and the claim is in it. grep -n "" # Pull the public AEVF spec — auditors do this without an account: curl -sS https://aegisagent.in/aevf/spec.md | head -20 # Download the reference bundle and verify it offline: curl -O https://aegisagent.in/aevf/reference-bundle-2026-06.json sha256sum reference-bundle-2026-06.json # → 8a6f09f65c374edf44c811dba8f146c8d79dab9ed74e3c49920be759951f20fc pip install 'aegis-aevf==1.1.0' aegis-verify --bundle reference-bundle-2026-06.json # → 6/6 PASS # Verify the running OpenAPI spec matches the reference. TOKEN=... curl -sS https://aegisagent.in/openapi.json | jq '.paths | keys | length' ``` Where this documentation and the running code disagree, the code wins. Please open an issue. ## Source layout ``` docs/ ├── AEVF/ # The open standard — spec, README, checklist, report, bundle ├── introduction/ # 5 pages ├── architecture/ # 7 pages ├── services/ # 18 service docs + index ├── ui/ # 34 UI docs + index ├── security/ # 7 pages ├── operations/ # 6 main + 3 runbooks ├── integrations/ # SDK + evidence-export adapter docs ├── api/ # 4 pages ├── SUMMARY.md # GitBook navigation └── README.md # this page ``` ## License * Aegis source code: **Apache License 2.0** * AEVF specification: **Apache License 2.0** (fork the spec, write your own verifier) * This documentation: **CC-BY-4.0** ## Links * **Live URL:** [aegisagent.in](https://aegisagent.in) — the single canonical URL (the legacy `ha.aegisagent.in` alias was retired on 2026-06-24) * **AEVF landing:** [aegisagent.in/aevf/](https://aegisagent.in/aevf/) * **Reference verifier on PyPI (1.1.0):** [`pip install 'aegis-aevf==1.1.0'`](https://pypi.org/project/aegis-aevf/) * **Framework SDKs (2026-06-24):** [`pip install 'aegis-anthropic==1.1.2'`](https://pypi.org/project/aegis-anthropic/) · [`pip install 'aegis-openai==1.1.2'`](https://pypi.org/project/aegis-openai/) · [`pip install 'aegis-langchain==1.1.3'`](https://pypi.org/project/aegis-langchain/) · [`pip install 'aegis-bedrock==1.1.3'`](https://pypi.org/project/aegis-bedrock/) * **Client onboarding (long-form):** [`guide.md`](https://github.com/Abhi-mishra998/aegis/tree/main/docs/guide.md) * **Repository:** [github.com/Abhi-mishra998/aegis](https://github.com/Abhi-mishra998/aegis) * **Security disclosures:** [`SECURITY.md`](https://github.com/Abhi-mishra998/aegis/blob/main/SECURITY.md) --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.aegisagent.in/readme.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.