Skip to main content

agentdevlabs

Python SDK for Agent Dev Labs Capture & Replay. wrap(client) records the calls your app makes through an OpenAI or Anthropic client. Your provider key stays with your app.

pip install agentdevlabs
import openai
from agentdevlabs import wrap

client = wrap(openai.OpenAI())

Set up with a coding agent

  1. Check that AGENT_DEV_LABS_TOKEN is set in .env. If it isn't, stop and ask the user for it. Never create a token or write one into code, and never commit .env. The SDK reads the process environment, not .env itself, so make sure the app loads it. In Python call load_dotenv() from python-dotenv before wrapping the client, or export the variable.
  2. Install agentdevlabs and the provider SDK (openai or anthropic) with the project's own package manager.
  3. Wrap the provider client where it is created: wrap(openai.OpenAI()) or wrap(anthropic.Anthropic()). Leave AGENT_DEV_LABS_MODE unset unless the user asks, because it outranks the devkit toggle.
  4. Run the app twice. The first call shows as captured and the second as replayed in the console's Logs; ask the user to look there, since you can't sign in.

Settings are environment variables only:

Variable Meaning
AGENT_DEV_LABS_TOKEN Your workspace token. Without it the client runs unwrapped.
AGENT_DEV_LABS_MODE auto, capture, replay or off, in any case. Overrides the server's mode. Any other value runs as auto, with one warning naming it.
AGENT_DEV_LABS_DISABLED 1, true, yes or on turns the SDK off; 0, false, no, off or empty leaves it on. Any other value turns it off, with one warning naming it. Beats every other setting.
AGENT_DEV_LABS_API_URL The API to talk to.
AGENT_DEV_LABS_REPLAY_SPEED How fast a replayed stream plays: realtime (the default), fast (10x), instant, or a number to divide the recorded pace by.
AGENT_DEV_LABS_GROUP Names a trace group: calls are captured into it and replays use only its recordings. Letters, digits, ., _, / and -, up to 100.
AGENT_DEV_LABS_CACHE_DIR Where replayed captures are cached, encrypted. Defaults to your OS cache directory.

Replay in CI

Record a group on your machine, then replay only that group in CI. A call with no recording in the group fails with an error naming the cache key and the group (or saying the group doesn't exist), so CI never reaches the provider.

The group name uses letters, digits, ., _, / and -, up to 100 characters; anything else fails at startup. AGENT_DEV_LABS_TOKEN in CI must be a replay-only service token, which an owner or admin creates in the console's Access tokens page.

AGENT_DEV_LABS_GROUP=checkout-flow AGENT_DEV_LABS_MODE=capture pytest   # once, locally
# .github/workflows/test.yml
name: test
on: [push, pull_request]
jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: astral-sh/setup-uv@v5
      - run: uv sync
      - run: uv run pytest
        env:
          AGENT_DEV_LABS_TOKEN: ${{ secrets.AGENT_DEV_LABS_TOKEN }}
          AGENT_DEV_LABS_GROUP: checkout-flow
          AGENT_DEV_LABS_MODE: replay

Project rules: a worked example

A travel agent's system prompt carries today's date, a session UUID and a request id, so Monday's recording never matches on Tuesday. In the console's Project rules page, three masks make the calls match:

  • [0-9]{4}-[0-9]{2}-[0-9]{2} masks the date.
  • [0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12} masks the session UUID.
  • req_[A-Za-z0-9]+ masks the request id.

A mask changes only what matching reads, so the recording still holds the original text. To see which mask a pair of calls needs, open a trace, choose "Why didn't this replay?" on step 1 and "Try this rule" on a suggestion.

A card number the user typed should never be uploaded, so add a redact pattern such as [0-9]{4} [0-9]{4} [0-9]{4} [0-9]{4}. The SDK replaces it with [redacted] before upload; matching still uses the original, so redacting doesn't stop replays. The page's "Try it" box shows both results on a sample request that never leaves the browser.

Calls already recorded keep the rules they were captured with: after saving a rule, capture them again.

License and privacy

Apache-2.0. The SDK sends nothing beyond the capture and replay events described above: no usage analytics or telemetry.

Versions

Versions follow semantic versioning. Before 1.0, a minor release (0.1 to 0.2) may break your code; the changelog says how, and the API supports only the latest 0.x minor. From 1.0, the API supports the previous major version for 12 months after a new major ships. See CHANGELOG.md.

Metadata

Release files for agentdevlabs 0.1.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for agentdevlabs 0.1.0
File Size Uploaded Provenance Actions
agentdevlabs-0.1.0.tar.gz 73.1 kB Without provenance Details for agentdevlabs-0.1.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for agentdevlabs 0.1.0
File Interpreter ABI Platform Actions
agentdevlabs-0.1.0-py3-none-any.whl Python 3 none any Details for agentdevlabs-0.1.0-py3-none-any.whl

Total release size: 118.1 kB

Release files / agentdevlabs-0.1.0.tar.gz

Download URL agentdevlabs-0.1.0.tar.gz
Size 73.1 kB
Tags Source
SHA-256 checksum
How to use checksums
1eb30f4b482bd44a00000f83108764230a073355ffe0c3f0485fa94359ccebb8
BLAKE2b-256 checksum
How to use checksums
d84777c3ecbb3652970b5151fa228da813126ecbc02916d438e495afa527e863
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Release files / agentdevlabs-0.1.0-py3-none-any.whl

Download URL agentdevlabs-0.1.0-py3-none-any.whl
Size 45.0 kB
Tags Python 3
SHA-256 checksum
How to use checksums
fe9ed14c0a1f12c1ff67fcf2f3a1e7112e5c04461c8f016bc919edb7f1560828
BLAKE2b-256 checksum
How to use checksums
53373fd895ad0b8809e791c8d4b676bf9dafa96e922a8f14718c2214a8398c30
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Security for agentdevlabs 0.1.0

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.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page