The AI swarm runtime
for complex tasks
devsper breaks any task into a DAG of agents, runs them in parallel, and synthesizes the results.
terminal
_
Everything you need to build with AI agents
Built for production. Designed for developers.
Parallel execution
Tasks run as a DAG — independent agents fire simultaneously.
Persistent memory
Agents remember across runs. Vector search over past context.
Tool ecosystem
100+ built-in tools. Extend with plugins from the registry.
Self-healing
Failed tasks are automatically diagnosed and retried with a new strategy.
Run intelligence
Cost tracking, bottleneck analysis, critical path visualization.
Distributed
Scale across nodes. Redis-backed, Rust worker, single binary to start.
Simple to start. Powerful at scale.
import devsper
result = devsper.run(
"Analyze this CSV and write a report with charts",
files=["sales_data.csv"],
model="claude-sonnet-4",
)
print(result.output) # report.md written
Share tools with the ecosystem
Publish plugins to the devsper registry. One command to install, one command to publish. pip-compatible.
$ devsper reg install devsper-plugin-browseruse
✓ Installed 3 tools: browse, click, extract_text
$ devsper run "Book me a flight to Tokyo next Friday"
⣾ Planning...