For Agent Builders
Describe what you want to build — a feature, an API, a full app — and 0spec generates a multi-agent flow that plans, delegates tasks, implements, runs QA loops, and reviews it end-to-end. Claude, Codex, and ACP agents collaborate with automatic dependency resolution and parallel execution.
Anthropic Claude via the Agent SDK. Opus, Sonnet, Haiku — with full tool use, MCP, and extended thinking.
OpenAI Codex via the Codex SDK. Threads, streaming responses, configurable reasoning effort.
Agent Communication Protocol. Connect Cursor, Windsurf, or any ACP-compatible coding agent as an engine.
Run Claude, Codex, and ACP agents in the same flow. Each stage picks the best engine for the job. Planning with Opus, implementation with Codex, review with both.
Independent stages run concurrently up to your plan limit. Fan-out after a shared planning stage, fan-in for consolidation. The scheduler handles dependency resolution.
Each stage declares which tools it needs. Planning stages get Read, Glob, Grep. Implementation stages get Edit, Write, Bash. Review stages get everything read-only.
Every run persists its full state. Resume from any failed stage, replay with different parameters, or inspect artifacts from completed stages.
Each session gets its own git worktree. Parallel agents write to isolated branches. No merge conflicts during execution.
Describe an app in natural language — or feed in a detailed spec — and 0spec generates a multi-agent flow that builds it end-to-end. Planning, task delegation, implementation, QA loops, and review stages are created automatically. Use --ref for mockups or docs, refine interactively with --chat, and save the result with --save.
Agents accumulate decisions, discoveries, and proven fixes across sessions. BM25 full-text search retrieves relevant memories per-stage.
Give each agent its own set of tools, system instructions, and MCP servers. A security reviewer gets SAST tools; a copywriter gets brand guidelines; a planner gets read-only access. Fine-grained control over what each agent can see and do.
Run any flow on a recurring schedule. Nightly QA passes, weekly refactor audits, daily research with open-source models — all unattended. Solo and Pro plans only.
Flows are TOML files. Each stage declares its team, tools, dependencies, and output format. The scheduler resolves the DAG and runs stages concurrently where possible.
[flow]
name = "plan-implement-review"
[[stages]]
id = "plan"
type = "par"
team = "architects"
prompt = "Analyze codebase. Produce implementation plan."
tools = ["Read", "Glob", "Grep"]
output = { format = "markdown", name = "plan.md" }
[[stages]]
id = "implement"
type = "seq"
team = "implementors"
depends_on = ["plan"]
input_from = ["plan"]
prompt = "Implement the plan. Write tests."
tools = ["Read", "Edit", "Write", "Bash"]
output = { format = "markdown", name = "summary.md" }
[[stages]]
id = "review"
type = "par"
team = "reviewers"
depends_on = ["implement"]
input_from = ["implement", "plan"]
prompt = "Review for correctness, security, coverage."
tools = ["Read", "Glob", "Grep", "Bash"]
output = { format = "markdown", name = "review.md" }0spec --query 'build auth module' --flow defaultRun the built-in default flow template
0spec --query 'add OAuth login' --ref mockup.png --ref spec.mdAuto-plan a flow with reference files for context
0spec --query 'build auth module' --chatInteractive flow refinement — chat with planner before execution
0spec --query 'build auth module' --save my-flowSave the planned flow as a reusable template
0spec --task 'review code for security' --run-agent reviewer-claudeOne-off task with a specific agent
0spec --resume ses_20260310_104437 --from implementResume from a failed stage
0spec --query 'audit deps' --cron '0 3 * * *'Schedule a flow to run nightly at 3 AM
0spec --statusShow subscription, agents, sessions