Published · 14h agoBuild3 min read
The agent harness is the product: DeepSeek ships a runtime that routes to its rivals
DeepSeek Harness is an MIT-licensed developer preview in which the model adapter, tool registry, session log and agent loop are all plugins. The interesting part is what that implies for evaluation.
Written for builders.See today for builders
What happened
- DeepSeek AI has released DeepSeek Harness, or dsh, as an open-source developer preview under the MIT license.
- Install it, point it at a provider, and you get file editing, shell access, web search, sub-agents and a web UI without further assembly.
- The underlying Cordis framework gives plugins a shared context plus services, typed events and reversible effects, drawn from a Peking University and DeepSeek paper.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraintIf context handling and error recovery sit in the runtime, model leaderboards under-determine how an agent will behave in week two, and evaluation budget has to move to the harness that will...
- exposureSandboxing and approval policy travelling inside a swappable bundle means an agent's blast radius is decided by whoever assembles the profile, not by the model's refusals.
- decisionBuyers now have to specify which runtime owns the session, tools and history, because the model slot behind it has become a configuration line rather than an architectural choice.
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
The surrounding machinery around a model is often called the agent harness: the runtime that sits between the model and the world and that actually determines how the agent behaves in practice.
ReportedView cited source - [2]
Two agents built on the same underlying model can behave completely differently depending on the harness wrapped around it: how it manages context, what tools it exposes, how it recovers from errors and how it schedules work.
ReportedView cited source - [3]
The shorthand of an LLM plus a system prompt plus a few tools works for demos but falls apart once an agent must run more than a few minutes, survive a restart, call a sub-agent, recover from a failed tool call, or let a human inspect what it did three hours ago.
ReportedView cited source - [4]
The source lists the infrastructure a long-running agent needs that has nothing to do with the model itself: a place to execute tools safely, a way to keep state across turns, a policy for what context the model sees on each call, a mechanism for delegating work to other agents, a sandbox to contain what the agent can touch, a way to recover from partial failures, and a replayable record of what happened.
ReportedView cited source - [6]
DeepSeek Harness (dsh) is an open-source project released by DeepSeek AI in developer preview under the MIT license.
ReportedView cited source - [7]
Installing dsh and pointing it at a model provider yields a working agent with file editing, shell access, web search, sub-agents and a web UI out of the box.
ReportedView cited source
Sources & coverage · 1 publisher
The reporting this story was synthesized from, earliest first. Every link goes to the original.
- dev.toPramoda Sahu15h agoDeepSeek Harness: What Happens When the Agent Runtime Becomes the Product
Cited in this coverage: DeepSeek Harness architecture documentation, quoted by dev.to
- dev.tomech.app1h agoDeepSeek Harness: When the Agent Runtime Becomes the Product