Pipelogic on autopilot.
Use your favourite LLM provider in combination with Pipelogic Agents to build AI systems via chat. Pipelogic Agents natively integrate with all major LLM providers via ppl CLI.
Build 20 times faster at production-ready quality.
Most infrastructure tools are built for humans first, with agents added as an afterthought. Pipelogic is agentic-native, built for agents as the main user.
- Manual Backend assembly, searching and wiring components by hand
- Human-in-the-loop for every deploy-test-undeploy cycle
- Fragile glue code to connect components and manage parameters
- Reading documentation, guessing at component compatibility
- Writing code, debugging Components, asking LLMs what went wrong
- Automated Backend assembly from a catalog of ever growing list of Components
- Full deploy-test-undeploy lifecycle with no human required
- Destructive operations refused at the CLI level — not classified at runtime
- AGENT.yml manifest on every component for confident selection decisions
- Automated Component creation via instructions or existing code repo
From zero to agent-ready in three commands.
Set up the CLI, authenticate your account, and choose an agent profile to begin building with PipeLogic.
ppl loginppl mode generalThen instruct your LLM to use ppl, and you're ready to go!
Safe development with task-specific permissions.
Not every agent should have access to every operation. Pipelogic profiles limit the CLI to exactly what each task requires, and nothing more.
ppl mode generalFull read/write across the platform. Default for broad-scope orchestrators managing multiple resource types, or before the task is fully known.
- Widest surface — good for initial orientation
- Navigate docs with ppl docs tree
- Switch to a scoped profile once task is known
ppl mode componentAuthoring, publishing, validation, and version management for custom Components. Keeps the surface on the build-and-release loop.
- Scaffold, compile, and release components
- Search the registry with --query
- Inspect versions and build logs
- Promote to released — human only
ppl mode backendCreate Backends, wire components, set parameters, bind files, deploy, and monitor. For agents assembling and operating AI infrastructure.
- List and query backends, runtimes, files, nodes
- Deploy, inspect logs, check runtime health
- Bind parameters, files, and credentials
- Overwrite released versions — human only
ppl mode ci-cdAutomated deploy/test/undeploy loops and health checks. Fully non-interactive — designed for pipelines, not interactive sessions.
- Non-interactive, no confirmation prompts
- Stable JSON output for scripting
- Per-invocation via --agent=ci-cd
ppl mode applicationCreate and deploy Applications; bind application endpoints to Backend roles. For agents handling frontend and endpoint binding tasks.
- Create and deploy Applications
- Bind application endpoints to Backend roles
- Inspect and monitor deployed Applications
- Promote Application to production — human only
Returning to human mode always requires a fresh ppl login — agents cannot silently swap surfaces.
Works with any coding agent.
Choose the LLM backend that fits your latency, privacy, and cost requirements. Too expensive, too slow, or too opinionated? Swap it anytime.
Claude Code
Codex CLI
Cursor
Amp
Goose
Qwen Code
Crush
Aider
GitHub Copilot CLI
OpenHands
Agents prepare. Humans commit.
The ppl CLI ships with built-in agent mode: a machine-readable execution layer for stable JSON output, profile-scoped permissions, and automatic refusal of destructive operations. No middleware. No wrapper APIs.
Machine-readable by default
Stable {ok, data} JSON for every command. Branch on IDs and status codes, not prose. Lists are capped and filterable, keeping context windows small and easy to parse.
Safe by construction
Destructive operations are refused at the CLI level, not classified by your agent at runtime. Agents prepare and test. Humans promote and release. The boundary is structural, not policy-based.
Token-efficient by design
20-record default caps, --query filters, and lean projections keep context windows small. Use --limit=N when a query could return more than 20 rows. Fetch by ID to retrieve the full record.
FAQ
Common questions about agent mode, profiles, output format, and safety boundaries.
A CLI execution mode designed for machine consumption — stable JSON output, profile-scoped permissions, no destructive operations, lean list projections. Activate with ppl mode <profile> or --agent=<profile> per command. It's not a separate product; it's a different execution mode of the same ppl CLI.
Use workspace secrets. Create and rotate them in an interactive human session. Your agent reads secret IDs from ppl secret list — values are never returned — and binds them to backend vertices using ppl backend change-parameter. At deploy time, the platform resolves the ID to the actual value. The raw credential never passes through your agent's context.
Yes. The ppl CLI runs wherever your infrastructure runs — private cloud, on-premise, or air-gapped. Workflows that use local inference components (Ollama, vLLM, SGLang) require no external connectivity. Hosted provider components (OpenAI, Anthropic, etc.) require outbound API access.
Almost. Agents cover the full build-deploy-monitor-test lifecycle. What they cannot do: promote a component to public release, execute destructive overwrites on released versions, or manage secrets (create, update, rotate, delete). Those require a human ppl login session by design. Agents prepare and test. Humans commit.
A machine-readable manifest that tells an agent when to pick a component, what behavioral constraints apply, and what setup is required. Built-in components already have them. For custom components, adding AGENT.yml makes your component agent-selectable. It's not required for deployment — it's a selection signal for intelligent agent-driven assembly.
No. Agents interact with Pipelogic through the ppl CLI. You write agent logic (prompts, orchestration) in whatever framework you use — the CLI is the stable, machine-readable surface your agent calls. Custom Components can be written in Python or C++ when your use case needs something the catalog doesn't cover.
Ready to Build?
Join the technical teams using Pipelogic to ship AI systems faster.