harness-sdk vs trueforge
Strands Agents: open-source Python and TypeScript SDK for building and running agent harnesses in your process — model-agnostic, with lifecycle controls, hooks, MCP, multi-agent and tracing. — versus — TrueFoundry's open agent harness: the runtime loop — model calls, MCP tools, SKILL.md packs, sandboxing, approvals, compaction — behind a chat UI, HTTP API, TypeScript SDK and embeddable UI.
Both package the full harness runtime — model calls, MCP tools, skills, approvals, compaction — behind an SDK; TrueForge adds a chat UI and HTTP API, Strands stays a library in two languages.
| harness-sdk | trueforge | |
|---|---|---|
| Stars | 8.8k | 6.1k |
| Forks | 1.4k | 495 |
| Language | Python | TypeScript |
| License | Apache-2.0 | MIT |
| Last activity | yesterday | 3 days ago |
| Topics | agents, orchestration | agents, orchestration |
| Curated connections | 5 | 6 |
harness-sdk — the curator's take
Pick it when you'd otherwise hand-roll an agent loop and want the grown-up parts — turn and token limits, cancellation, hooks to intercept any step, guardrails, structured output — without a hosted control plane. `create_harness()` gives a tuned default agent; drop to the SDK to own the loop. Strongest on AWS (Bedrock first-class), though Anthropic, OpenAI, Gemini and Ollama work. Want an explicit graph of steps and durable checkpoints? LangGraph models that more directly.
trueforge — the curator's take
Take it when the agent logic is the easy part and you don't want to hand-build streaming, session persistence, tool auth, sandboxing and an approval UI: YAML catalogs configure models, MCP servers, skills and a sandbox once, then agents pick from what you connected. Context engineering is first-class — subagents, deferred tool loading, Code Mode, large-result offloading, compaction — and it scales from `npx` on SQLite to Postgres+Redis via Docker Compose or Helm. Caveats: local mode has no login and is explicitly not for anything internet-facing, sandboxing means Daytona today, and it's a harness rather than a graph framework — if you need bespoke control flow per step, a lower-level runtime fits better.