langchain-ai/deepagents deepagents==0.7.0b1

on GitHub

Rolls up alphas 0.7.0a1–0.7.0a8 plus some additional changes not captured in an a9 release (see footer's Git log summary).

Breaking Changes

• Agents created with create_deep_agent no longer include TodoListMiddleware by default, so the write_todos tool, todos state channel, and todo-planning prompt are absent unless restored with middleware=[TodoListMiddleware()]. (The OpenAI Codex harness profile still opts in automatically.) (#4929)
• Default agent prompts are lean: the authored base prompt starts empty, tool-usage prose that duplicates tool schemas is trimmed, and the previous authored base prompt remains available as DEFAULT_AGENT_PROMPT. BASE_AGENT_PROMPT is deprecated and no longer exported at the package top level. (#4859) (#4979)
• Removed deprecated backend compatibility shims. Callers must pass concrete BackendProtocol instances (not factories), configure StoreBackend with an explicit namespace, and use the current ls / glob / grep / ReadResult APIs. Removed compatibility symbols include BackendFactory, BACKEND_TYPES, FileFormat, and Unset, plus the deprecated runtime parameters on StateBackend and StoreBackend. New files store string FileData.content; older list[str] content remains readable and is converted when overwritten or edited. (#4541)
• ls and glob empty-result output is now No files found instead of []. Callers that parse empty results should update those checks.
• read_file no longer renders raw text with a fixed-width cat -n-style line-number gutter and tab separator. Line and continuation markers are dynamically aligned and separated from source content by two spaces. Callers that parse raw tool output should update those parsers. (#4561)

Features

• Custom middleware passed to create_deep_agent(..., middleware=[...]) can replace a default middleware instance when .name matches, so defaults such as SummarizationMiddleware can be overridden without also excluding the built-in instance. (#4251)
• Filesystem backends and middleware support delete, including recursive deletes, with a user-facing delete tool and clear unsupported/not-found errors when a backend cannot perform the operation.
• write_file can overwrite existing files, so agents can replace generated artifacts or apply full-file edits without an intermediate delete. (#4109)
• FilesystemMiddleware accepts a tool allowlist so callers can expose only selected built-in filesystem tools while keeping read_file available and leaving custom user tools unaffected. Omitted built-in tools are non-executable. (#4325) (#4698)
• The deepagents[aws] extra includes Bedrock prompt caching support. (#4108)
• Automatically enable Fireworks prompt-cache session affinity when a compatible langchain-fireworks installation is available. (#4598)
• grep and glob bound large or slow trees by returning valid partial results with a truncated flag instead of hanging or discarding work on timeout; glob also supports brace expansion such as *.{py,md} on supported backends. (#4063)
• Agent-facing grep defaults to a 1,000-match cap, streams local ripgrep output, and stops once the cap is reached. Direct FilesystemBackend.grep() callers can request surrounding lines with keyword-only context_lines. (#4570) (#4706)
• Paginated read_file responses report the returned source-line range, total and remaining line counts, and the next offset, including when reads are shortened by sandbox or middleware limits. (#4540)
• Optional video frame extraction for read_file when media dependencies are installed. Video files such as .mp4, .mov, .webm, and .mkv are sampled into JPEG frames, with offset and limit interpreted as seconds. (#4094)
• Sandbox large-result offload for sandbox.execute captures oversized command output directly inside the sandbox artifact path to reduce round trips. (#4230)
• Built-in NVIDIA adds a built-in Nemotron 3 Ultra harness profile. We also expand with NIM app-origin attribution. (#4455) (#4192)

Bug Fixes

• Keep fields marked with PrivateStateAttr, including fields declared through create_deep_agent(state_schema=...), out of subagent inputs and returned parent-state updates. (#4587)
• Preserve ContextT through the create_deep_agent(..., middleware=[...]) type annotation so type checkers accept context-aware middleware when a matching context_schema is passed. (#4055)
• Accept YAML list values as well as comma-separated strings for skill allowed-tools frontmatter, and make skill truncation warnings actionable with field name, path, length, configured limit, and impact. (#4140) (#4141)
• Align filesystem instructions with the tools that remain after allowlist and backend-capability filtering, so agents no longer reference hidden grep/glob tools or prohibit equivalent shell search when dedicated search tools are unavailable. (#4920) (#4921)
• Propagate default-backend failures from CompositeBackend.ls("/") and CompositeBackend.als("/") instead of returning successful route-only listings. (#4925)
• Correct CompositeBackend.glob / CompositeBackend.aglob routing so explicit default-backend paths such as /tools do not also return files from routed backends such as /memories. (#4531)
• Constrain sandbox glob and slash-pattern grep searches to their declared search root by treating leading / as search-root-relative, rejecting .. traversal segments, and filtering symlink-resolved matches outside the root. (#4588)
• Unify grep(..., glob=...) include-glob semantics across filesystem and in-memory backends: basename patterns like *.py match at any depth, and slash-containing patterns like src/**/*.py match relative paths consistently. (#3936)
• Make grep's literal-only contract clearer in tool descriptions and no-match hints, route slash-containing sandbox include-globs correctly, and shorten default search timeouts so bad patterns and huge trees return guidance faster. (#4168)
• Align sandbox delete behavior with other backends by returning not-found errors for missing paths, and avoid over-blocking unrelated sibling deletes when deny rules use glob patterns. (#4321)
• Emit max_iterations_reached as the terminal RubricMiddleware status when the iteration cap is exhausted, instead of a final needs_revision event that will not loop. (#4406)
• Improve rubric grader failure diagnostics with configured model, structured-output strategy, and integer HTTP status when available. (#4938) (#4967)
• Handle missing async subagent URLs consistently in check_async_task and cancel_async_task. (#3967)

Thanks to our community contributors: @kavishkartha05 (Twitter, LinkedIn), @nv-kasikritc

Internal maintainers: @nick-hollon-lc, @ccurme, @davibinboi, @mdrxy

Released by: @mdrxy

Released from beta/deepagents-0-7-0b1 at commit a631817