Skip to content

reference.md

Latest commit

 

History

History
169 lines (126 loc) · 4.78 KB

reference.md

File metadata and controls

169 lines (126 loc) · 4.78 KB

AnyMind Reference

Entry Points

CLI

The anymind console script is backed by ../src/anymind/cli/main.py.

Important options:

  • --agent
  • --config / -c
  • --model
  • --provider
  • --temperature
  • --tools-policy
  • --tools-enabled / --no-tools
  • --thread-id
  • --query / -q
  • --log-level

Without --query, the CLI runs in interactive mode.

API

The FastAPI app lives in ../src/anymind/api/app.py.

Endpoint Purpose
GET /health Liveness check.
POST /agents/run Synchronous agent execution.
POST /jobs Async execution backed by JobManager.
GET /jobs/{job_id} Job status and result retrieval.
POST /jobs/{job_id}/pause Pause a running job.
POST /jobs/{job_id}/resume Resume a paused job.
POST /jobs/{job_id}/cancel Cancel a job.

Agents

Agent Implementation Notes
research_agent research_director_agent.py Delegates probe questions to other runtimes and synthesizes the final answer.
sop_agent sop_agent.py Expects a JSON SOP payload or @path to a JSON file.
aiot_agent aiot_agent.py Brain/worker iterative loop with validated JSON checkpoints.
giot_agent giot_agent.py Parallel agent loop with temperature spread and facilitator-style convergence.
agot_agent agot_agent.py Adaptive graph planning and worker-pool execution.
got_agent got_agent.py Graph-of-thought search with reflection and verification.

Configuration Discovery

Model config

Search order:

  1. AM_MODEL_CONFIG
  2. ./model.json
  3. ./config/model.json
  4. repo-root/config/model.json
  5. ~/.config/anymind/model.json

Important fields in ModelConfig:

  • model
  • model_provider
  • model_parameters
  • thread_id
  • tools_enabled
  • tools_policy
  • budget_tokens
  • search
  • state_dir
  • checkpoint
  • cache
  • agent-specific sections such as aiot, giot, agot, got, sop, and research_director

MCP config

Search order:

  1. AM_MCP_CONFIG
  2. ./mcp_servers.json
  3. ./config/mcp_servers.json
  4. repo-root/config/mcp_servers.json
  5. ~/.config/anymind/mcp_servers.json

The MCP registry resolves relative paths against the working directory used to build the session.

Tool Policies

Policy Behavior
auto Expose all configured tools.
planner Ask the model to choose a subset of tools for the current request.
confirm Expose all tools, but require a terminal confirmation before execution.
never Disable tool use.

When the model provider is ollama and the config says auto, the session factory upgrades the policy to planner.

Built-in Local MCP Tools

The bundled local MCP server is ../src/anymind/tools/mcp_local_tools.py. It exposes:

  • current_time
  • pdf_extract_text
  • internet_search when Kagi and Scrapfly credentials are available

These can be used through the local MCP registry just like external servers.

Credential requirements:

  • internet_search is only registered when Kagi plus Scrapfly credentials are available
  • current_time does not require external API credentials
  • pdf_extract_text does not require Kagi or Scrapfly credentials

Persistence and Caching

  • Checkpoints default to SQLite through AsyncSqliteSaver when available.
  • If the SQLite checkpoint dependency is unavailable, the code falls back to in-memory checkpoints.
  • checkpoint.backend=redis is not currently implemented and raises an error.
  • Usage data can be stored in Redis via ANYMIND_USAGE_REDIS_URL.
  • Tool-result caching can wrap MCP calls when cache.redis_url is set in the model config.

Environment Variables

Config and state

  • AM_MODEL_CONFIG
  • AM_MCP_CONFIG
  • AM_STATE_DIR

Logging and display

  • ANYMIND_LOG_PATH
  • ANYMIND_LOG_DIR
  • ANYMIND_LOG_LEVEL
  • ANYMIND_LOG_RUN_ID
  • ANYMIND_EVIDENCE_MAX_CHARS
  • ANYMIND_EVIDENCE_ITEM_MAX_CHARS

Tool and embedding passthrough

These may be forwarded into MCP server environments by the registry:

  • KAGI_API_KEY
  • KAGI_API_ENDPOINT
  • SCRAPFLY_API_KEY
  • SCRAPFLY_API_KEY_SECRET_ARN
  • SCRAPFLY_COUNTRY
  • SCRAPFLY_FORMAT
  • AWS_ACCESS_KEY_ID
  • AWS_SECRET_ACCESS_KEY
  • AWS_SESSION_TOKEN
  • AWS_PROFILE
  • AWS_REGION
  • AWS_DEFAULT_REGION
  • ONNX_MODEL_PATH
  • ONNX_TOKENIZER_PATH
  • ONNX_MAX_LENGTH
  • ONNX_EMBED_BATCH_SIZE
  • PDF_ONNX_MODEL_PATH
  • PDF_ONNX_TOKENIZER_PATH
  • PDF_ONNX_MAX_LENGTH

For adapters of this repo, the minimum search-related setup is:

  • KAGI_API_KEY or search.kagi_api_key in the model config
  • SCRAPFLY_API_KEY or SCRAPFLY_API_KEY_SECRET_ARN

Without those values, internet_search is not registered by the bundled local MCP server.