Add Redis Iris

Inspect, install, route. Do not rewrite the agent.

1. Find the agent

Locate:

• The turn: where the model is called (invoke, Runner.run, generate_content, agent.run, kickoff, chat.completions.create, messages.create).
• Tool registration: where tools are defined and handed to the model.
• User and session ids: where the request's user id and thread or session id live.
• More than one loop (a hand-written loop and a framework, or two frameworks)? Create the product objects once in a module with no framework imports, wire each loop at its own seam, and convert one loop at a time. Do not pick the loop for the user.

2. Detect the framework

Search imports:

Import	Framework
langgraph, langchain, langchain_core	LangGraph / LangChain
agents (package openai-agents)	OpenAI Agents SDK
google.adk	Google ADK
pydantic_ai	Pydantic AI
crewai	CrewAI
only openai, anthropic, or another model client	Plain loop

Note the version from the lockfile. Product skills look up that version's API over MCP.

3. Find what it lacks

Symptom in the code	Product	Skill
No long-term memory, InMemoryStore, checkpointer only, memory in a dict or local DB	Agent Memory	add-agent-memory
Already on Redis Agent Memory through its own REST or MCP client, hand-written adapters, or model-facing memory tools	Agent Memory	add-agent-memory, "Already on Redis Agent Memory"
Policy or procedures hardcoded in the prompt, tool usage guidance in the prompt, or a vector store over policy docs	Agent Playbooks	add-agent-playbooks
A semantic response cache: model answers stored and re-served to later users, curated by hand	Agent Playbooks	add-agent-playbooks, "Replacing a response cache"
Several MCP servers configured one by one in the agent, each with its own secret, or too many tool schemas in the prompt	Context Gateway	add-context-gateway
A keyword if/else router, or an LLM call just to pick a route	Intent Router	add-intent-router
No per-turn tracing: no way to see which LLM calls, tools and retrievals one answer used, or what it cost	Lens	add-lens

If the user named a product, use that. If unclear, ask once, listing what you found.

Lens goes next to any product, or on its own: it traces the turn, and every Iris product traces itself inside it. Offer it alongside whatever else you add.

4. Install

• uv.lock or [tool.uv] present: uv add redis-iris.
• Poetry: poetry add redis-iris.
• Otherwise add redis-iris to the project's requirements file.

If resolution fails on pre-releases, retry with uv add --prerelease=allow redis-iris. Confirm: uv run python -c "import redis_iris; print(redis_iris.__version__)".

5. Route

Follow each chosen product skill. More than one product: each skill adds its own calls. Optionally build them from one config with Iris.from_env() (iris.agent_memory(), iris.agent_playbooks(), iris.context_gateway(), iris.intent_router()).

Ask once whether the services run on Redis Cloud or a self-managed suite, then write the project's iris.toml profile. Only the iris CLI reads iris.toml; the app reads env vars. It holds URLs only: a token in TOML makes the CLI exit 2. Select the profile with IRIS_CONFIG=./iris.toml and IRIS_PROFILE=cloud or IRIS_PROFILE=suite.

Redis Cloud. The person fills .env from the console: IRIS_<PRODUCT>_URL, its ID and _API_KEY per product.

[profiles.cloud]
kind = "cloud"
console_url = "https://cloud.redis.io"

[profiles.cloud.agent_memory]
url = "https://gcp-us-east4.memory.redis.io"  # the data-plane URL from the console

Self-managed suite. The operator's tokens go in .env.admin, never the app's .env; service create --write-env .env writes the app's .env.

[profiles.suite]
kind = "self-managed"

[profiles.suite.agent_memory]
service_url = "https://memory-admin.example.com"  # control plane: the CLI calls it for service create and check
url = "https://memory.example.com"  # data plane: what the agent calls
identity_url = "https://memory-identity.example.com"  # Identity Service that mints the agent's keys

Add one table per product the app uses: agent_memory, agent_playbooks, context_gateway, intent_router. [profiles.suite.context_gateway] has no url: service create --write-env writes the Gateway's endpoint. A suite with one shared Identity Service sets [profiles.suite.identity] url once instead of each identity_url. IRIS_<PRODUCT>_SERVICE_URL, IRIS_<PRODUCT>_URL and IRIS_<PRODUCT>_IDENTITY_URL override the profile.

Rules

• Keep the user's framework, graph, and prompts. Smallest diff.
• Never guess framework APIs from memory. Look them up over MCP (see each product skill).
• Only use APIs in redis_iris. redis_iris.integrations.langgraph.attach (extra redis-iris[langgraph]) and redis_iris.integrations.adk.attach (no extra) exist; there is no IrisTool, iris.tools(), or iris.turn().
• The SDK raises on a missing setting or an outage. Ask whether a product failure may fail the turn; most production agents catch it and continue with an empty block.
• Never commit secrets.