A
Agent-S
85.4
12,994 stars · Apache-2.0
AI Agents
deepseek-harness leads by 12.0 · clear
A
85.4
12,994 stars · Apache-2.0
B · leads
97.4
282,805 stars · MIT
Agent-S versus deepseek-harness is a AI Agents comparison assembled from public repository metrics, the license on file, and whether each product can be self-hosted. Agent-S is written mainly in Python; deepseek-harness is written mainly in TypeScript. The composite score is 85.4 for Agent-S and 97.4 for deepseek-harness.
deepseek-harness has 282,805 GitHub stars to Agent-S's 12,994, a gap of 269,811 (21.76×). Forks are 1,519 for Agent-S and 35,041 for deepseek-harness. On the log scale used inside this pair, adoption is 76 versus 100, and the fork community is 70 versus 100.
Push history: Agent-S was last pushed on 13 November 2026 (3 days ago), and deepseek-harness was last pushed on 13 November 2026 (4 days ago). Open issues stand at 82 for Agent-S (0.006 per star) and 0 for deepseek-harness (0.000 per star). Issue-health scores are 84 and 96; activity scores are 100 and 92.
Agent-S is listed as Apache-2.0, which this catalog treats as a permissive grant. A company can usually ship it inside a commercial product and keep its own application code private, provided it keeps the copyright notice. deepseek-harness is listed as MIT, which this catalog treats as a permissive grant. A company can usually ship it inside a commercial product and keep its own application code private, provided it keeps the copyright notice. License scores in this pair are 96 for Agent-S and 96 for deepseek-harness. This is a procurement signal, not legal advice.
In AI Agents, both Agent-S (12,994 stars) and deepseek-harness (282,805 stars) are marked self-hostable. Deploy scores are 100 and 100. Running either one means you own upgrades, backups, and the license duties that follow a modified copy. Pricing labels on file: Agent-S is open source, deepseek-harness is open source.
The lead is split. Agent-S is furthest ahead on recent activity (100 to 92), while deepseek-harness is furthest ahead on fork community (100 to 70). Choose the dimension a rollout cannot compromise on.
deepseek-harness finishes ahead at 97.4 against Agent-S at 85.4, a 12.0 point clear margin under the published weights.
Scores are computed inside this pair only. Method and weights.
Bars are scaled within the row. Higher is stronger. Open issues are a raw count, so that row has no bar.
| Signal | Agent-S | deepseek-harness |
|---|---|---|
| GitHub stars | 12,994 | 282,805 |
| Forks | 1,519 | 35,041 |
| Open issues | 82 | 0 |
| Adoption | 76 | 100 |
| Activity | 100 | 92 |
| Community | 70 | 100 |
| Issue health | 84 | 96 |
| License | 96 | 96 |
| Self-hosting | 100 | 100 |
| Composite | 85.4 | 97.4 |
Agent S: the state-of-the-art open source computer use agents
DeepSeek Harness: Everything is a Plugin.
deepseek-harness has more GitHub stars: 282,805 versus 12,994 for Agent-S. The gap is 269,811 stars.
Agent-S was pushed more recently: 13 November 2026 versus 13 November 2026 for deepseek-harness. That is 1 days apart.
Yes. This catalog marks both Agent-S and deepseek-harness as self-hostable. Confirm each project's deploy docs before you plan a migration.
Agent-S is listed as Apache-2.0, which this catalog treats as a permissive grant. A company can usually ship it inside a commercial product and keep its own application code private, provided it keeps the copyright notice. deepseek-harness is listed as MIT, which this catalog treats as a permissive grant. A company can usually ship it inside a commercial product and keep its own application code private, provided it keeps the copyright notice. Read both grants before you embed either codebase in a hosted product. This is not legal advice.
deepseek-harness scores 97.4 and Agent-S scores 85.4. This page calls that a clear lead for deepseek-harness.