feat: move from mozilla any-agent to langchain deepagents (backport #44)

[mergify]

---

This is an automatic backport of pull request #44 done by [Mergify](https://mergify.com).

[greptile-apps]

Confidence Score: 5/5

The migration is architecturally sound and the single non-blocking observation (asyncio.run inside a sync tool) is safe under LangGraph's current thread-pool dispatch model.

This is a large but well-contained framework swap. The harness profile lock, explicit VFS deny permissions, and typed subagent response schemas are all in place. The only observation worth noting is the asyncio.run() + lazy import in extract_document_data, which works today but relies on LangGraph always dispatching sync tools through run_in_executor rather than inline. No data-loss, auth bypass, or schema/migration concerns are present.

ask_alyf/ask_alyf/toolset.py — specifically the extract_document_data method's asyncio.run() bridge pattern.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant FE as Frontend
    participant API as api.py
    participant RM as run_message()
    participant Runner as ask_alyfAgentRunner
    participant DA as Deep Agents<br/>(create_deep_agent)
    participant TS as ask_alyfToolset
    participant SA as Subagents<br/>(source-code-analyzer / document-planner)
    participant VFS as CompositeBackend<br/>(/workspace/ /source/ /attachments/)

    FE->>API: POST /ask
    API->>RM: run_message(conversation_name, message, mode, request_context, history)
    RM->>Runner: ask_alyfAgentRunner(runtime)
    Runner->>Runner: _ensure_ask_alyf_harness_profile() [once per process, locked]
    Runner->>Runner: build_chat_model(settings)
    Runner->>VFS: build_ask_alyf_backend(app_roots)
    Runner->>DA: create_deep_agent(model, tools, system_prompt, backend, subagents, permissions)
    Runner->>Runner: build_input_messages(history + new HumanMessage)
    Runner->>DA: "agent.invoke({messages})"
    DA->>TS: tool call (get_list / get_meta / insert / ...)
    TS-->>DA: result
    DA->>SA: task tool → subagent invocation
    SA->>VFS: read-only ops (/source/ or /attachments/)
    VFS-->>SA: FileData / LsResult / GrepResult
    SA-->>DA: SourceCodeAnalysisResult / DocumentPlannerResult
    DA-->>Runner: "{messages: [...AIMessage]}"
    Runner-->>RM: "{response, pending_operations, document_extractions, attached_files}"
    RM-->>API: dict
    API-->>FE: JSON response

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant FE as Frontend
    participant API as api.py
    participant RM as run_message()
    participant Runner as ask_alyfAgentRunner
    participant DA as Deep Agents<br/>(create_deep_agent)
    participant TS as ask_alyfToolset
    participant SA as Subagents<br/>(source-code-analyzer / document-planner)
    participant VFS as CompositeBackend<br/>(/workspace/ /source/ /attachments/)

    FE->>API: POST /ask
    API->>RM: run_message(conversation_name, message, mode, request_context, history)
    RM->>Runner: ask_alyfAgentRunner(runtime)
    Runner->>Runner: _ensure_ask_alyf_harness_profile() [once per process, locked]
    Runner->>Runner: build_chat_model(settings)
    Runner->>VFS: build_ask_alyf_backend(app_roots)
    Runner->>DA: create_deep_agent(model, tools, system_prompt, backend, subagents, permissions)
    Runner->>Runner: build_input_messages(history + new HumanMessage)
    Runner->>DA: "agent.invoke({messages})"
    DA->>TS: tool call (get_list / get_meta / insert / ...)
    TS-->>DA: result
    DA->>SA: task tool → subagent invocation
    SA->>VFS: read-only ops (/source/ or /attachments/)
    VFS-->>SA: FileData / LsResult / GrepResult
    SA-->>DA: SourceCodeAnalysisResult / DocumentPlannerResult
    DA-->>Runner: "{messages: [...AIMessage]}"
    Runner-->>RM: "{response, pending_operations, document_extractions, attached_files}"
    RM-->>API: dict
    API-->>FE: JSON response

Loading

_{Reviews (3): Last reviewed commit: "refactor(toolset): remove orphaned code-..." | Re-trigger Greptile}

[barredterra]

Cherry-picked the follow-up fixes from the develop PR onto this backport branch: harness profile registration lock, subagent VFS tooling clarification, and explicit langchain/langchain-core/langgraph dependency declarations.

[barredterra]

Cherry-picked a follow-up cleanup onto this backport branch: removed the five orphaned code-search toolset wrapper methods (search_code, read_code_file, ls, find, grep) that were no longer registered as agent tools after the move to the source-code-analyzer subagent. The underlying tools.* functions remain.

[barredterra]

Note: the cleanup commit on this backport branch is also being PR'd against develop separately (the develop PR for the prior fixes merged before the cleanup could be added to it), so develop and the backports stay in sync.