Redis Iris
One package for every Iris product: an opinionated Python API per product, and the iris CLI to set them all up.
Install
Requires Python 3.11+. uv is the recommended installer.
uv add redis-iris # library, in your project
uv tool install redis-iris # iris CLI, isolated
iris --help
pipx or pip
pipx install redis-iris
python -m pip install redis-iris # inside an active virtual environment
uv projects with the fastapi-sse extra. uv takes a pre-release only when your own project
names it, and the extra pins fastapi-redis-sdk to a pre-release, so uv add fails with
hint: pre-releases weren't enabled. Restate the pin in your pyproject.toml:
[tool.uv]
constraint-dependencies = [
"fastapi-redis-sdk==0.10.0b1",
]
For a LangGraph agent, install redis-iris[langgraph].
From a source checkout, uv tool install --force . installs or replaces the iris tool, and
python -m redis_iris runs the CLI wherever the package is installed.
Alpha releases may change commands and output contracts. Pin a version in automation and check the release notes before upgrading.
A2A integrations (beta)
Agent Mesh is in beta. Features and behavior are subject to change.
The optional a2a integration provides RedisTaskStore and RedisTaskEventStream
for hosting A2A agents across API replicas, a producer heartbeat, and HTTP+JSON
streaming routes that resume from Last-Event-ID. It requires Redis 8.6.2+ with JSON. They live
under redis_iris.agent_mesh.a2a and accept application-owned async Redis clients.
The stack
Iris is a suite of products. Use one, or all of them.
| Product | What your agent gets | Instead of |
|---|---|---|
| Agent Memory | Ordered session history, summarized automatically. Durable memories extracted in the background, typed to your domain and steered away from sensitive data. Recall across sessions, scoped to one customer plus shared team knowledge. Erase a customer in one call. | an in-process store, a hand-built extraction chain, regex scrubbing |
| Agent Playbooks | Answers only from published, versioned FAQs and Skills, scored. Rollback and withdrawal. Proposes an FAQ for human review when nothing covers the question. | a vector store of unreviewed policy docs |
| Context Gateway | One MCP endpoint and one agent key for every MCP server and Context Surface. Per-tool cache, and compression: large results as handles the agent pages through. | a config entry and a secret per MCP server in every agent |
| Intent Router | The user's prompt matched to one of the intents you defined, with its score and the dispatch target you attached. | a prompt classifier you tune by hand, or an LLM call per routing decision |
| Lens | One trace per turn: which memories and documents entered the prompt, tokens, cost. | guessing why the agent said that |
The CLI
iris <product> <command> creates services, mints keys, and manages content. Shared options go
before the product name.
iris agent-memory service create --from service.yaml --write-env .env
iris agent-memory service key <store-id> --actions read --name inspector --write .env.inspector
iris agent-memory memories search "how does this customer want answers" --owner cust-1042 --prompt
iris agent-playbooks extraction bind <store-id> --idle-seconds 60 --enable # FAQ proposals from sessions
iris agent-playbooks faqs validate ./faqs # offline
iris agent-playbooks faqs sync ./faqs --publish changed
iris --json --no-input agent-playbooks faqs plan ./faqs # unattended, one JSON value on stdout
iris --profile production agent-playbooks health
iris --json context-gateway service list # Context Gateway control plane
iris intent-router service create --name support --provider openai --model text-embedding-3-small \
--dimensions 1536 --write-env .env
iris identity keys list --product intent # agent keys, for every product
Settings resolve from command options, environment variables, the selected TOML profile, shared
TOML settings, then defaults. API keys belong in environment variables. Exit codes are 0 for
success, 1 for operational failures, and 2 for invalid input or configuration. Help, version,
completion, and local validation work without credentials or a network connection.
Upgrading to Agent Playbooks
The product is now Agent Playbooks; a Playbook is one instance. This follows the core SDK. Old
environment variables and the old TOML table are still read as fallbacks, so existing .env
files and TOML keep working; the CLI writes only the new names. iris agent-playbook stops with
a pointer to the new group, and the Python names have no aliases.
| Before | Now |
|---|---|
iris agent-playbook |
iris agent-playbooks |
IRIS_AGENT_PLAYBOOK_ID |
IRIS_AGENT_PLAYBOOKS_PLAYBOOK_ID |
IRIS_AGENT_PLAYBOOK_* (URL, API_KEY, CONTENT_API_KEY, SERVICE_URL, ...) |
IRIS_AGENT_PLAYBOOKS_* |
TOML [agent_playbook], [profiles.<name>.agent_playbook] |
[agent_playbooks], [profiles.<name>.agent_playbooks] |
redis_iris.agent_playbook, AgentPlaybook, AgentPlaybookSettings |
redis_iris.agent_playbooks, AgentPlaybooks, AgentPlaybooksSettings |
IrisSettings.agent_playbook, Iris.agent_playbook() |
IrisSettings.agent_playbooks, Iris.agent_playbooks() |
A new variable wins over its old name; set it empty to ignore the old one. The old TOML table is read only when the new one is absent.
License
Apache License 2.0.
Metadata
Release files for redis-iris 1.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| redis_iris-1.1.0.tar.gz | 1.2 MB | Without provenance | Details for redis_iris-1.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| redis_iris-1.1.0-py3-none-any.whl | Python 3 | none | any | Details for redis_iris-1.1.0-py3-none-any.whl |
Total release size: 1.6 MB
Release files / redis_iris-1.1.0.tar.gz
| Download URL | redis_iris-1.1.0.tar.gz |
|---|---|
| Size | 1.2 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
797b13a8aedd058c64778b7a627c5ce209e1418ff8c6352f81c05158db94cd67
|
|
BLAKE2b-256 checksum How to use checksums |
1367cf245c5897f94bf767b420afedbb03aaebe96fe6d0a940101f24932cea92
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Release files / redis_iris-1.1.0-py3-none-any.whl
| Download URL | redis_iris-1.1.0-py3-none-any.whl |
|---|---|
| Size | 409.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
64c2f3c487effed50f873d22e06be7cbd00b5b8ef18a546263648fd5f894e4ba
|
|
BLAKE2b-256 checksum How to use checksums |
1c937231ea687ecb05736d5a74947db01f21f3bfbf35f4a6eef6b7013ea57f1b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Security for redis-iris 1.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.