Mission SDK — the ReDevOps Runtime front door

ReDevOps Runtime is a production runtime layer for AI applications: planning, context, governed execution, verification, replay, governance, and runtime optimization. The Mission SDK (redevops-mission) is its application-facing front door.

Works with your existing LangGraph, LangChain, custom agents, tools, models, and infrastructure. Keep your agent framework. Stop rebuilding the production runtime around every app.

Quickstart · Architecture · Examples · For coding agents

Category: Agent Automation / Agent Runtime Infrastructure — not a vector database, RAG library, generic workflow engine, or LLM framework.

Install

pip install redevops-mission          # the SDK + its pinned agentic-os runtime; exposes the `rdo` command

From a clone (until the package is on your index): pip install -e .. Optional adoption extras: redevops-mission[langgraph], [langchain], [telemetry], [full]. The default install needs no provider key and no network to run the minimal example. Check the environment any time with rdo doctor.

Minimal working example

from redevops_mission import (
    MissionProgram, Operator, capability, step, template, run_program, explain, export_bundle, replay_bundle,
)

@template("hello_mission")                        # 1. outcomes + their dependency shape
def hello_mission(mission_id):
    return [step("greeting_ready", need="produce a greeting for the user")]

OPERATORS = [Operator("greeter", [                # 2. capabilities that provide them (wrap YOUR logic)
    capability("greet.hello", handler=lambda inputs: {"greeting": "hello from the runtime"},
               provides=["greeting_ready"]),
])]

PROGRAM = MissionProgram.from_template("hello_mission", goal="Greet the user", grants=[])  # 3. one artifact

result = run_program(PROGRAM, OPERATORS)          # execute
exp    = explain(PROGRAM, OPERATORS)              # why this plan
replay = replay_bundle(export_bundle(PROGRAM, OPERATORS), OPERATORS)   # reproduce the sealed run
print(result.succeeded, replay.consistent)        # -> True True

Runnable, offline, and CI-tested at examples/00_minimal/main.py:

python examples/00_minimal/main.py
# run     : state=succeeded succeeded=True nodes=1
# explain : goal='Greet the user', 1 node(s), first=greet.hello -> greeting_ready
# replay  : recorded=succeeded replayed=succeeded consistent=True integrity_ok=True
# OK — mission executed, explained, replayed, and verified.

What ReDevOps is (and is not)

ReDevOps Runtime	Does not replace
runtime layer beneath applications	LangGraph
context / evidence planning	LangChain
governed mission execution	your business logic
verification / replay	your model provider
runtime telemetry / governance seams	your infrastructure
provider / runtime optimization	your application UI

It runs beneath your agent. The single most useful guide is Add ReDevOps to an existing agent without replacing it — the correct integration wraps your existing call in a capability; it does not rewrite your agent.

What the public tests prove

• installation works from a clean environment;
• the minimal mission executes offline and deterministically;
• an existing agent can be wrapped without a rewrite;
• replay reproduces the sealed plan and terminal state;
• a case bundle verifies its own integrity without re-running;
• the security / telemetry plane is opt-in, not required for basic use.

The full map is docs/public-test-matrix.md. Run it: python -m pytest -q (or a guarantee group, e.g. python -m pytest tests/adoption -q).

Operate a mission (rdo CLI)

rdo doctor                                        # environment + live minimal-mission check
rdo mission validate examples/revenue_rescue/mission.py   # static + compile checks (no execution)
rdo mission explain  examples/revenue_rescue/mission.py   # the compiled physical graph
rdo mission simulate examples/revenue_rescue/mission.py   # dry-run cost/latency/approvals/success
rdo mission run      examples/revenue_rescue/mission.py --approve   # execute on the local profile
rdo mission bundle   examples/revenue_rescue/mission.py --out run.json   # seal a replayable bundle
rdo mission replay   examples/revenue_rescue/mission.py run.json         # reproduce it

Full verb list and the M3 status: QUICKSTART.md.

Public AGPL stack vs enterprise extensions

The public AGPL stack is the open runtime contract and reference implementation — canonical contracts, public runtime interfaces, and context/planning/execution/replay capabilities as actually shipped, with the public examples and tests. It is not a crippled demo.

Enterprise extensions (separate, private repos) add production security enforcement, telemetry bridges, deployment adapters, secret-store plugins, and long-haul storage. They are optional and never required for public SDK use.

Repository role & map

mission-sdk is the public front door — onboarding, composition, examples — and links to the canonical repos rather than forking their implementations. See docs/repo-map.md for which repo owns each subsystem (runtime-contracts, context-runtime, discovery-runtime, agentic-os, redevops-rag), COMPATIBILITY.md for pinned versions, and PARITY.md for the SDK↔runtime contract check.

Developing against a local runtime

pip install -e . resolves the pinned agentic-os from git. For development against a local runtime checkout, set AGENTIC_OS_SRC=/path/to/agentic-os — a loud, opt-in override (unset ⇒ a clear ImportError, so the SDK is never silently satisfied by an unknown checkout). See ARCHITECTURE.md and CONTRIBUTING if present.