mirror of
https://github.com/bytedance/deer-flow.git
synced 2026-09-14 16:08:41 +00:00
13f0a7f263
* feat(extensions): let an out-of-tree extension observe what the agent did
DeerFlow's extension system can contribute middleware, services and routes,
but an extension cannot answer basic questions about a run without reaching
into host internals. Several of the facts it would need are destroyed by the
operations that produce them:
* The middleware chain injects and rewrites a lot of context — date
reminders, recalled memory, compaction summaries, durable-context data,
image payloads, activated skill bodies. Downstream, none of it is
attributable: at the model-call boundary an injected HumanMessage is
indistinguishable from the user's own, and anything wanting to tell them
apart has to pattern-match prompt wording, which breaks on the next copy
edit.
* Two runs of "the same agent" are only comparable if the chain enforced the
same limits, prompts and thresholds. Recovering that from outside means
reading private attributes and guessing which of them change behaviour — a
guess that rots silently as middlewares gain fields.
* The lead-agent factory resolves a model after runtime overrides, renders a
prompt, filters tools through authorization and composes a stack, all
inside one synchronous call, and none of it survives: a middleware sees its
neighbours but not the prompt, the run worker sees a graph but not what
went into it.
* Summarization is destructive by design. N messages leave the context and
one summary enters it; afterwards only the summary exists, so "which
messages became this?" is not reconstructible.
This adds seven neutral facilities so those facts are recorded where they are
still true, and releases the contract package as 0.2.0.
Message provenance
Producers stamp `deerflow_content_kind` / `deerflow_producer_kind` onto the
messages they inject or rewrite. Stamping is unconditional — a fact whose
presence depends on whether an observer is installed is not a fact — and the
keys are server-owned, so provenance cannot be forged from a request.
Middleware self-description
Twelve middlewares declare their own behaviour-affecting parameters through
a duck-typed `release_policy_parameters()`. Long text is hashed rather than
embedded: a declaration is an identity, not a copy of the prompt.
Agent assembly descriptor
`assemble_lead_agent()` returns the graph plus a descriptor whose fingerprint
answers "did anything about this agent change between these two runs?".
`make_lead_agent()` keeps its graph-only signature — it is the LangGraph
Server ABI declared in langgraph.json. Tools and skills are sorted before
hashing because their assembly order is incidental; middlewares are not,
because stack order decides what wraps what. Host build identity is reported
but excluded from the fingerprint, so a redeploy does not invalidate every
agent's identity.
Context compaction observation
Summarization emits the content hashes of the messages it is about to remove
joined to the summary that replaced them. Content is the only identity
available at that seam: the summary does not become a message, and what later
projects it into a request renders it bounded and escaped rather than
verbatim.
Neutral policy, transform and MCP-source facts
Guardrail decisions are published to runtime context under a `__`-prefixed
key; result-rewriting middlewares append a declared, ordered transform trail;
MCP tools carry their credential-free logical origin.
Extension route identity
Contributed routes are session-authenticated and cannot opt out, but
"logged in" and "administrator" are different questions. Extensions get a
neutral projection of the caller rather than the host's auth context, and
`require_admin` fails closed when identity cannot be determined.
Extension-owned tables
An extension that persists data owns its own MetaData and migration chain, so
its tables are absent from Base.metadata and `alembic revision --autogenerate`
proposes dropping them. Extensions declare a table prefix, which is rejected
at registration if it would shadow a host table.
The contract package stays dependency-free and imports no host code; every new
Protocol method has a default so later additions remain additive. The loader's
pre-1.0 rule requires an exact major.minor match, so extensions written against
0.1 are now refused at startup with an actionable install hint rather than
loading into a host that implements a different surface.
uv.lock records the contract package's new version, so `uv sync --locked` still
resolves on a fresh checkout.
* fix(backend): sort gateway service imports
112 lines
4.9 KiB
TOML
112 lines
4.9 KiB
TOML
[project]
|
|
name = "deer-flow"
|
|
version = "2.1.0"
|
|
description = "LangGraph-based AI agent system with sandbox execution capabilities"
|
|
readme = "README.md"
|
|
requires-python = ">=3.12"
|
|
dependencies = [
|
|
"deerflow-harness",
|
|
# Direct dependency on purpose, even though deerflow-harness already pulls
|
|
# it in: app/gateway/app.py and app/gateway/services.py import the public
|
|
# contract package themselves (the extension principal resolver, and the
|
|
# provenance key set the state route strips). Those are the app layer's own
|
|
# imports of a package whose whole point is a stable public surface, so the
|
|
# app declares them rather than relying on the harness to keep supplying it.
|
|
"deerflow-extension-api",
|
|
"fastapi>=0.115.0",
|
|
"httpx>=0.28.0",
|
|
"python-multipart>=0.0.31",
|
|
"sse-starlette>=2.1.0",
|
|
# Direct dependency on purpose, even though FastAPI already pulls it in:
|
|
# app/gateway/request_path.py imports the private
|
|
# `starlette._utils.get_route_path` so the auth and CSRF predicates
|
|
# classify the exact string Starlette's router matches on. A private
|
|
# import is the safest option here because it fails loudly (ImportError at
|
|
# startup) instead of silently drifting from the dispatcher at a security
|
|
# boundary -- but it does mean a Starlette bump is a security-relevant
|
|
# change. Declaring and bounding it here makes that bump visible in the
|
|
# diff; tests/test_gateway_request_path.py pins the agreement itself.
|
|
"starlette>=1.3.1,<2",
|
|
"uvicorn[standard]>=0.34.0",
|
|
"lark-oapi>=1.4.0",
|
|
"slack-sdk>=3.33.0",
|
|
"python-telegram-bot>=21.0",
|
|
"langgraph-sdk>=0.1.51",
|
|
"markdown-to-mrkdwn>=0.3.1",
|
|
"wecom-aibot-python-sdk>=0.1.6",
|
|
"dingtalk-stream>=0.24.3",
|
|
"bcrypt>=4.0.0",
|
|
"pyjwt>=2.13.0",
|
|
"email-validator>=2.0.0",
|
|
"e2b-code-interpreter>=2.8.1",
|
|
]
|
|
|
|
[project.optional-dependencies]
|
|
postgres = ["deerflow-harness[postgres]"]
|
|
redis = ["deerflow-harness[redis]"]
|
|
discord = ["discord.py>=2.7.0"]
|
|
buzz = ["coincurve>=20.0.0"]
|
|
monocle = ["deerflow-harness[monocle]"]
|
|
browser = ["deerflow-harness[browser]"]
|
|
memory-zh = ["deerflow-harness[memory-zh]"]
|
|
|
|
[dependency-groups]
|
|
# Managed extension packages are added here by `deerflow extensions install`.
|
|
# Keeping them separate from development tooling lets every startup mode sync
|
|
# the same locked runtime set without promoting those packages to core deps.
|
|
extensions = []
|
|
dev = [
|
|
"blockbuster>=1.5.26,<1.6",
|
|
"hypothesis>=6.100,<7",
|
|
"jsonschema>=4.26.0",
|
|
"prompt-toolkit>=3.0.0",
|
|
"pytest>=9.0.3",
|
|
"pytest-asyncio>=1.3.0",
|
|
"ruff>=0.14.11",
|
|
# Monocle tracer (also the deerflow-harness[monocle] extra); kept in the dev
|
|
# group so the tracing tests can import it without forcing it onto installs.
|
|
"monocle_apptrace>=0.8.8",
|
|
# redis is an optional runtime extra (deerflow-harness[redis]); pin it in the
|
|
# dev group so the stream-bridge tests can always import/exercise the redis
|
|
# bridge without forcing it onto production installs.
|
|
"redis>=5.0.0",
|
|
# TUI runtime dep (also declared as the deerflow-harness[tui] extra); kept in
|
|
# the dev group so the terminal workbench can be run and tested locally / in CI.
|
|
"textual>=0.80",
|
|
]
|
|
|
|
[tool.pytest.ini_options]
|
|
markers = [
|
|
"no_auto_user: disable the conftest autouse contextvar fixture for this test",
|
|
"allow_blocking_io: opt out of the strict Blockbuster gate in tests/blocking_io/",
|
|
"integration: tests that require an external service (e.g. Redis); skipped when unavailable",
|
|
"live: tests that call real external APIs and require explicit opt-in",
|
|
]
|
|
|
|
[tool.uv]
|
|
index-url = "https://pypi.org/simple"
|
|
default-groups = ["dev", "extensions"]
|
|
# langgraph-sdk 0.4.2 (pulled in by langgraph 1.2.9 for DeltaChannel) pins
|
|
# `websockets<16,>=14`, silently downgrading websockets 16.0 -> 15.0.1. The
|
|
# pin is not grounded in any API incompatibility: websockets 16's only
|
|
# breaking change is requiring Python >=3.10 (we require >=3.12), the sdk
|
|
# only imports `websockets.asyncio.client`/`websockets.exceptions` (both
|
|
# 16-compatible), and DeerFlow never uses the sdk's WebSocket transport
|
|
# (httpx/SSE only). DeerFlow does have one direct consumer of its own: the Buzz
|
|
# channel (`app/channels/buzz.py`) imports the top-level `websockets` package and
|
|
# calls `websockets.connect()`, which in 16.0 is the same asyncio client the sdk
|
|
# uses, re-exported at the package root -- so it is covered by the same
|
|
# compatibility argument. Pin the exact pre-upgrade 16.0 for that plus the IM
|
|
# channel integrations (dingtalk-stream, python-telegram-bot, etc.) that ran on
|
|
# it before. Remove once langgraph-sdk relaxes the pin upstream. Note: enabling
|
|
# the `openai[realtime]` or `slack-sdk[optional]` extras would conflict (they
|
|
# also cap websockets<16).
|
|
override-dependencies = ["websockets==16.0"]
|
|
|
|
[tool.uv.workspace]
|
|
members = ["packages/harness", "packages/extension-api"]
|
|
|
|
[tool.uv.sources]
|
|
deerflow-harness = { workspace = true }
|
|
deerflow-extension-api = { workspace = true }
|