mirror of https://github.com/bytedance/deer-flow.git synced 2026-09-25 22:16:19 +00:00

History

fix(memory): report malformed backend_config values by key name (#5555 )

* fix(memory): report non-mapping Honcho backend_config values as ValueError

failure_policy, workspace_overrides and user_peer_overrides were read with a
falsy-only `or {}` fallback, so a truthy non-mapping (a bare string, a YAML
list) reached .get/.items and escaped as a bare AttributeError from inside
backend construction. Route all three through one _mapping helper that keeps
falsy values meaning "unset" and names the offending key as a ValueError, the
posture the mem0 and OpenViking backends already take.

* fix(memory): name the Honcho numeric knobs that cannot be cast

timeout_seconds / connect_timeout_seconds / message_char_limit /
max_injection_chars still reached float() / int() with a YAML null or a
mapping, so the operator got a TypeError naming neither the key nor the
config file. Narrow all four through one _number helper that keeps falsy
values meaning "unset" the way the sibling api_key / storage_path scalars
already do, and reports a value that cannot be cast as a ValueError on the
key. Numeric strings keep parsing, since that is what float/int accept.

* fix(memory): report non-numeric backend_config knobs by name in mem0 and OpenViking

mem0 casts top_k, score_threshold, max_injection_chars and timeout_seconds, and
OpenViking casts timeout_seconds, max_seen_message_ids, retrieval.top_k and
retrieval.max_injection_chars, straight through int()/float(). A knob written
without a value in YAML therefore escapes as "TypeError: int() argument must be
a string..." from inside backend construction, naming neither which knob nor
which file is wrong, and a non-numeric value escapes as the equally anonymous
"could not convert string to float".

Both now resolve numeric knobs through a helper that keeps a value-less key at
its default, the way the same dicts already treat failure_policy and
allow_insecure_http, and turns an uncastable value into a ValueError that names
the knob. OpenViking's score_threshold keeps None as a meaningful value rather
than a default to fall back to.

2026-09-20 16:43:54 +08:00

__init__.py

feat(memory): add mem0 HTTP memory backend (#4528 )

2026-07-29 07:11:20 +08:00

client.py

feat(memory): add mem0 HTTP memory backend (#4528 )

2026-07-29 07:11:20 +08:00

config.py

fix(memory): report malformed backend_config values by key name (#5555 )

2026-09-20 16:43:54 +08:00

mem0_manager.py

feat(memory): add opt-in relevance-aware retrieval ranking (#5251 )

2026-09-18 11:32:04 +08:00

message_filtering.py

fix(messages): drop legacy <uploaded_files> tag handling (#4826 )

2026-09-01 15:26:33 +08:00

README.md

feat(memory): add mem0 HTTP memory backend (#4528 )

2026-07-29 07:11:20 +08:00

README.md

mem0 memory backend

Uses mem0 (Platform hosted API, or any API-compatible self-hosted server) as DeerFlow's memory store. Fully stateless in-process: dedup, fact extraction, and storage are server-side, so it is safe for multi-worker Gateway deployments.

Configuration

memory:
  enabled: true
  injection_enabled: true
  manager_class: mem0
  mode: middleware            # or "tool"
  backend_config:
    api_key_env: MEM0_API_KEY          # key read from env, never in config.yaml
    base_url: https://api.mem0.ai      # or your self-hosted mem0 server
    allow_insecure_http: false         # true only for trusted local HTTP dev
    top_k: 8
    score_threshold: 0.1
    max_injection_chars: 12000
    timeout_seconds: 10
    startup_policy: fail_fast          # fail_fast | tolerate
    failure_policy:
      read: fail_open                  # fail_open | fail_closed
      write: log_and_drop              # log_and_drop | raise

Set the key in the environment: export MEM0_API_KEY=...

base_url must use HTTPS because every request carries the API key. For a trusted local-development server that only exposes HTTP, opt in explicitly with allow_insecure_http: true; do not use that setting across an untrusted network.

Identity mapping

DeerFlow	mem0
`user_id`	`user_id`
`agent_name`	`agent_id`
`thread_id`	`run_id`

Limitations

mode: middleware recall is query-less (the get_context contract carries no query): the bucket's most recent top_k memories are injected. For query-aware semantic recall use mode: tool.
mode: tool retains the passive per-turn write middleware for this backend, because mem0 extracts and deduplicates facts from conversations through add(). The agent still gains query-aware memory_search, while new conversations continue accumulating memory even though fact CRUD is not available.
Fact CRUD, import_memory, and Settings-page memory editing are not implemented (gateway returns 501). DeerMem remains the default backend.
No migration of existing DeerMem data.
log_and_drop write policy is at-most-once: a failed write is dropped.
memory_add/memory_update/memory_delete are backed by fact CRUD, which this backend does not implement; they return a clear unsupported-operation error. Conversation writes still happen through the retained middleware.

Async execution and failure behavior

The mem0 HTTP client is synchronous for compatibility with the MemoryManager contract. DeerFlow offloads it at every async boundary: the async middleware uses the manager's a* methods, and Gateway memory routes run sync management calls in worker threads. A slow mem0 request therefore does not block unrelated ASGI handlers or SSE heartbeats.

failure_policy.read: fail_open logs a recall failure and continues without new memory context. fail_closed propagates the backend error through prompt construction and aborts the run instead of silently degrading.