{"id":48625319,"url":"https://github.com/fboiero/argentor","last_synced_at":"2026-05-15T04:01:54.407Z","repository":{"id":339878486,"uuid":"1163644183","full_name":"fboiero/Argentor","owner":"fboiero","description":"Secure multi-agent AI framework in Rust — WASM sandbox, 50+ skills, 14 LLM providers, agent intelligence, compliance modules","archived":false,"fork":false,"pushed_at":"2026-05-04T20:56:05.000Z","size":9087,"stargazers_count":1,"open_issues_count":6,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2026-05-09T04:46:53.579Z","etag":null,"topics":["agent-framework","ai-agents","compliance","llm","mcp","multi-agent","rust","security","wasm"],"latest_commit_sha":null,"homepage":"https://fboiero.github.io/Argentor","language":"Rust","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"agpl-3.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/fboiero.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":"CODE_OF_CONDUCT.md","threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":"docs/ROADMAP_POST_V1.1.md","authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-02-21T23:32:57.000Z","updated_at":"2026-05-04T20:56:09.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/fboiero/Argentor","commit_stats":null,"previous_names":["fboiero/agentor","fboiero/argentor"],"tags_count":6,"template":false,"template_full_name":null,"purl":"pkg:github/fboiero/Argentor","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/fboiero%2FArgentor","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/fboiero%2FArgentor/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/fboiero%2FArgentor/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/fboiero%2FArgentor/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/fboiero","download_url":"https://codeload.github.com/fboiero/Argentor/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/fboiero%2FArgentor/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":33053144,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-13T13:14:54.681Z","status":"online","status_checked_at":"2026-05-15T02:00:06.351Z","response_time":103,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent-framework","ai-agents","compliance","llm","mcp","multi-agent","rust","security","wasm"],"created_at":"2026-04-09T04:30:54.498Z","updated_at":"2026-05-15T04:01:54.401Z","avatar_url":"https://github.com/fboiero.png","language":"Rust","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Argentor\n\n**Secure, high-performance multi-agent AI framework in Rust — WASM sandboxed plugins, MCP native, compliance-ready.**\n\n[![CI](https://github.com/fboiero/Argentor/actions/workflows/ci.yml/badge.svg)](https://github.com/fboiero/Argentor/actions/workflows/ci.yml)\n[![License: AGPL-3.0](https://img.shields.io/badge/License-AGPL--3.0--only-blue.svg)](https://www.gnu.org/licenses/agpl-3.0)\n[![Rust](https://img.shields.io/badge/Rust-1.80%2B-orange.svg)](https://www.rust-lang.org)\n[![Crates](https://img.shields.io/badge/Crates-17-informational.svg)]()\n[![PyPI](https://img.shields.io/badge/PyPI-argentor-blue.svg)](https://pypi.org/project/argentor/)\n\n---\n\n## Quick Start\n\n```bash\ngit clone https://github.com/fboiero/Argentor.git\ncd Argentor\ncargo run --example hello_world\n```\n\n---\n\n## Why Argentor?\n\n- **Security-first** — WASM sandbox + capability-based permissions on every skill; shell injection, null-byte, URL-decode, and NFKC path attacks blocked in the guardrails pipeline\n- **Rust performance** — ~2 ms framework overhead per turn vs 11–55 ms for Python-based competitors (N=10 paired, p \u003c 0.0001)\n- **WASM sandboxing** — plugins run in wasmtime/WASI isolation; ed25519 signature verification and static analysis before load\n- **MCP native** — full client/server/proxy with centralized control plane, credential vault, and token pool\n\n---\n\n## Benchmarks\n\nArgentor vs LangChain, CrewAI, PydanticAI, and Claude SDK across 6 dimensions (reproducible from `benchmarks/`):\n\n| Metric | Argentor | Best Competitor | Advantage |\n|--------|----------|-----------------|-----------|\n| Framework latency | ~2 ms | 11–55 ms | **~11x faster** |\n| Token cost (50-tool workload) | baseline | 7.9x more | **7.9x cheaper** |\n| Adversarial prompt blocking | 58.3% | 0% | **+58 pp security** |\n\nFull report: [docs/BENCHMARK_SYNTHESIS.md](docs/BENCHMARK_SYNTHESIS.md) — includes sensitivity analysis and honest losses.\n\n---\n\n## Architecture\n\n17 crates, strict workspace lints (`unwrap_used`, `expect_used` → warn in CI):\n\n| Crate | Description |\n|-------|-------------|\n| `argentor-core` | Core types, errors, event bus, metrics export, correlation context |\n| `argentor-security` | Capabilities, RBAC, rate limiting, audit log (rotating, background writer), TLS/mTLS, JWT, encrypted store, alerts, SLA tracking |\n| `argentor-session` | Session management, `FileSessionStore`, persistence |\n| `argentor-skills` | Skill trait, `SkillRegistry` (concurrent), WASM sandbox runtime, vetting pipeline, marketplace download/install/cache |\n| `argentor-agent` | AgentRunner, 14 LLM backends, Bedrock (SigV4, feature-gated), failover, streaming, circuit breaker, response cache (concurrent LRU+TTL), token budget, adaptive context compaction |\n| `argentor-channels` | Slack, Discord, Telegram, Webchat adapters |\n| `argentor-gateway` | HTTP/WebSocket gateway, auth, webhooks, Prometheus metrics, observability dashboard (HTML/JS, no deps), OpenAPI |\n| `argentor-builtins` | shell, file I/O, HTTP, memory, browser, Docker, code generation, 50+ universal skills |\n| `argentor-memory` | Vector memory, hybrid search (BM25 + embeddings), query expansion, JSONL persistence |\n| `argentor-mcp` | MCP client/server/proxy, proxy orchestrator, credential vault, token pool |\n| `argentor-orchestrator` | Multi-agent engine, TaskQueue with DAG, AgentMonitor, DeploymentManager |\n| `argentor-compliance` | GDPR, ISO 27001, ISO 42001, DPGA compliance modules |\n| `argentor-a2a` | Google A2A protocol: AgentCard, A2AServer, A2AClient, JSON-RPC 2.0 |\n| `argentor-tee` | TEE provider stubs (AWS Nitro, Intel SGX, AMD SEV-SNP), attestation verifier |\n| `argentor-cloud` | Multi-tenant managed runtime, TenantManager, QuotaEnforcer, 4-tier billing |\n| `argentor-python` | PyO3 Rust-to-Python bindings (maturin build, excluded from workspace tests) |\n| `argentor-cli` | CLI binary (`serve`, `deploy`, `agents`, `health`, `skill list`) |\n\n---\n\n## Features\n\n### Security\n- WASM sandboxed plugins (wasmtime + WASI) — capability-based permissions per skill\n- Guardrails: shell injection, base64-decode attacks, unicode normalization (NFKC), null-byte, URL-decode path attacks\n- SSRF prevention — blocks localhost, link-local, and private ranges\n- Path traversal protection — canonicalization + blocklist + null-byte detection\n- PII detection (Luhn, SSN, email, phone) with redaction\n- Prompt injection blocking — 23+ pattern signatures\n- RBAC policy engine, JWT/mTLS, encrypted credential store (AES-256-GCM)\n- Audit log rotation with background writer thread\n\n### Intelligence\n- Token budget per session + adaptive context compaction (4 strategies)\n- Extended Thinking Mode (Quick/Standard/Deep/Exhaustive) with confidence scoring\n- Self-Critique Loop (Reflexion pattern across 6 quality dimensions)\n- Dynamic Tool Discovery (TF-IDF + keyword hybrid, ~98% token reduction)\n- Learning Feedback Loop — tool selector improves from execution outcomes\n- 6-phase competitive benchmark suite (vs LangChain, CrewAI, PydanticAI, Claude SDK)\n\n### Developer Experience\n- 5 runnable examples (see `examples/`)\n- Workflow DSL — TOML-based agent workflows, no Rust code required\n- Tool Builder — 3-line tool definitions\n- Config hot-reload via file watcher (500ms debounce)\n- CLI REPL with 12 commands for interactive agent debugging\n- Observability dashboard at `/dashboard` — pure HTML/JS, no build step\n- Enterprise readiness report at `/api/v1/enterprise/readiness` — runtime score, active checks, available controls, and next actions\n\n### Integrations\n- 14 LLM providers: Claude, OpenAI, Gemini, OpenRouter, Groq, Ollama, Mistral, xAI, Azure OpenAI, Cerebras, Together, DeepSeek, Cohere, HuggingFace\n- AWS Bedrock backend — real SigV4 signing, feature-gated (`bedrock` feature flag)\n- 5,800+ tool integrations via MCP ecosystem\n- Channel adapters: Slack, Discord, Telegram, Webchat\n- Google A2A interoperability protocol\n- Python SDK (pip-installable) + TypeScript SDK (npm)\n\n---\n\n## Getting Started\n\n### Prerequisites\n\n- Rust 1.80+ (`rustup update stable`)\n- An API key from any supported LLM provider\n\n### Run Examples\n\n```bash\n# Hello world agent\ncargo run --example hello_world\n\n# Full DevOps pipeline demo (no API key needed)\ncargo run -p argentor-cli --example demo_pipeline\n\n# See all examples\nls examples/\n```\n\n### Configure\n\n```toml\n# argentor.toml\n[model]\nprovider = \"claude\"\nmodel_id = \"claude-sonnet-4-20250514\"\napi_key = \"${ANTHROPIC_API_KEY}\"\nmax_tokens = 4096\nmax_turns = 20\n\n[server]\nhost = \"0.0.0.0\"\nport = 3000\n```\n\n### Run the Server\n\n```bash\ncargo run --bin argentor -- serve\ncargo run --bin argentor -- skill list\n```\n\n### Run Tests\n\n```bash\ncargo test -p argentor-core -p argentor-agent -p argentor-security\ncargo test --workspace\ncargo clippy --workspace   # 0 warnings\ncargo fmt --all -- --check\n```\n\n---\n\n## Python SDK\n\n```bash\npip install -e python/\n```\n\n```python\nfrom argentor import ArgentorClient\n\nclient = ArgentorClient(base_url=\"http://localhost:3000\")\nresult = client.run_task(\"my_skill\", \"input text here\")\nprint(result[\"response\"])\n```\n\nAsync support and 24 typed models included. For native Rust bindings: `pip install maturin \u0026\u0026 cd crates/argentor-python \u0026\u0026 maturin develop`.\n\n---\n\n## Observability Dashboard\n\n```bash\ncargo run --bin argentor -- serve\n# Open http://localhost:3000/dashboard\n```\n\nPure HTML/JS — no build step, dark-themed SPA with deployment management, agent catalog, and health monitoring.\n\n---\n\n## Contributing\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines, and [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md).\n\nBug reports, feature requests, and pull requests are welcome. For security issues, see [SECURITY.md](SECURITY.md).\n\n---\n\n## License\n\n**GNU Affero General Public License v3.0 only** — see [LICENSE](LICENSE).\n\nAGPL-3.0-only applies to the entire Argentor family. No MIT, no Apache dual-licensing.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Ffboiero%2Fargentor","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Ffboiero%2Fargentor","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Ffboiero%2Fargentor/lists"}