# Astrid

Astrid is a portable, capability-secure operating system for composable software.

Repository: https://github.com/astrid-runtime/astrid
Canonical: https://ross.abutalabs.com/products/astrid-os
Language: Rust
License: Apache-2.0
License Family: permissive
Topics: agent-runtime, ai-agents, capability-security, llm, mcp, microkernel, rust, sandbox, wasm, webassembly
Last push: 2026-08-26T23:58:25+00:00
Link (homepage): https://github.com/astrid-runtime/astrid

## Health v2 (maintenance only)
Score: 80/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 93, longevity 14
- inputs: {"age_days": 199, "days_push": 7, "days_rel": 44, "gap_med": 2.5, "n_releases_24m": 17}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 10273, forks 135 (observed 2026-08-28T04:10:41.429018+00:00)

## What it is
Astrid is a portable, capability-secure operating system for composable software, where every component runs as a sealed WebAssembly capsule with no ambient authority. Its small Rust microkernel routes events, enforces cryptographic ed25519 capability grants, runs the Wasmtime sandbox, and records audit trails, making it a runtime-level security boundary for AI agents.

## Use cases
- run AI agents in a sandboxed runtime with no ambient file or network access
- grant agents explicit, revocable capabilities for files, hosts, and tools
- compose agent components as WASM capsules with a WIT-typed ABI
- prevent prompt injection and jailbreaks from escalating privileges
- audit every external effect an agent attempts
- build and distribute custom agent distros as signed archives
- host MCP tools behind capability-checked host calls

## When to choose
- you need OS-grade security boundaries for untrusted agent code rather than prompt-based trust
- you want cryptographic, revocable capability grants scoped to resource patterns
- you want to compose agent components in WASM with explicit authority
- you need an audit trail of all agent side effects

## When to avoid
- you just want a simple agent framework with prompt-level guardrails and no sandboxing
- you need a mature 1.0 product - it is pre-1.0 with low crate download counts
- you need deep integration with a specific model provider or agent loop, which Astrid deliberately avoids
- you cannot adopt Rust/WASM tooling in your stack

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, security, mcp, llm-inference, cli, wasm, cryptography
- domain: security, large-language-models, developer-tools, self-hosted
- platform: rust, cli, cross-platform
- tags: capability-security, wasm-sandbox, agent-runtime, microkernel, ed25519-grants, wasmtime, wit-abi, capsules, audit-trail, distro, sandbox, ai-agents, macos, linux

## Member repositories
- astrid-runtime/astrid (main) score 80
- astrid-runtime/sdk-js (sdk) score 74
- astrid-runtime/book (docs) score 55
- astrid-runtime/handbook (docs) score 54
- astrid-runtime/sdk-rust (sdk) score 76
- astrid-runtime/rfcs (docs) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:41.429018+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:19:21.551642+00:00, confidence not recorded.
  - readme: https://github.com/astrid-runtime/astrid (fetched 2026-08-28T04:10:41.429018+00:00, sha 87125e3f67fd)
  - homepage: https://github.com/astrid-runtime/astrid (fetched 2026-08-29T08:19:33.644264+00:00, sha d2ded29ec0bf)
  - registry_crates: https://crates.io/api/v1/crates/astrid (fetched 2026-08-29T08:19:33.654128+00:00, sha df4fe01a04a2)
- Data as of 2026-08-30T08:39:29.467469+00:00.