Agent-first Runtime
Run agents as managed daemons with session context, tool access, and reachable surfaces through CLI, SDK, or Console.
Features
Downcity gives builders the reusable runtime layer that agent products keep rebuilding.
Run agents as managed daemons with session context, tool access, and reachable surfaces through CLI, SDK, or Console.
Sessions, logs, runtime state, and task artifacts keep agent behavior visible and continuous across interactions.
One runtime for models, tools, tasks, memory, services, and operations across many products and entry points.
Add capabilities through plugins, skills, City APIs, and SDK extensions without forking the whole backend.
Runtime boundaries, permissions, audit logs, and operator surfaces keep execution inspectable and recoverable.
Embed agents into CLI workflows, browser apps, desktop consoles, or third-party platforms with the same backend.
Use cases
Let agents collect sources, summarize findings, and maintain a living knowledge base across sessions.
Turn structured data and long-form notes into consistent drafts with traceable source references.
Answer questions, explain code, and guide users through complex documentation without losing thread.
Index repo docs, notes, and API references so agents answer from your actual content instead of generic training data.
Draft, review, and publish content through agent-assisted pipelines with human approval at every checkpoint.
Documents, embeddings, and structured context live close to the agent so answers are grounded and auditable.
01Models, system prompts, sessions, and memory form a reusable brain that can be attached to any surface.
02Tools, plugins, tasks, and permissions turn model outputs into real, governed operations inside your workspace.
03features.repo.description
Projects keep Skills and runtime assets; Agent, model, and Plugin config is managed by one global database.
Project Asset
The only versioned Agent capability asset in a project
---
name: market-research
description: Research markets with verified sources
---
# Market Research
1. Gather primary and secondary sources.
2. Record citations for every material claim.
3. Compare competitors on one shared rubric.
4. Save the final report under ./reports/.
Builder Decision
Close one real agent product loop on a reusable runtime, then expand it into more workflows and product surfaces.