Skip to content

Cog Arch Gps

cog_arch_gps — General Problem Solver cognitive architecture.

Cluster: Experience & Autonomy | Type: component | MCP Tools: 29

Overview

General Problem Solver (GPS) cognitive architecture that finds operator sequences reducing the difference between a current state and a goal state using means-ends analysis. Delegates 29 operations to an inner GPSMCPBlock; operators are stored as JSON and retrieved for BFS/DFS search, with LLM augmentation available for operator generation and explanation.

When to use:

  • Solving well-defined planning problems by symbolic means-ends analysis over operator libraries
  • Generating step-by-step operator sequences for tasks expressible as state-transition problems
  • Augmenting LLM planning with a structured search algorithm that guarantees goal reachability

Launch readiness caveat

Component-level logic and regression tests support cog_arch_gps as useful for pilot planning workflows: high-level goals map into solver goals, stored problems can be solved by problem_id, inline and stored operators are supported, malformed JSON returns structured GPS errors, and step solving handles recursive preconditions. This does not by itself prove the full first-user MCP experience. Before presenting this component as launch-ready, run a clean-machine MCP install/runtime smoke test through the actual client, confirm the 29 GPS tools are discoverable, solve one stored-problem workflow end to end, and verify Docker/runtime packaging for the target deployment.

Example:

from mvp.cog_arch_gps import CogArchGPSBlock, GPSInput

block = CogArchGPSBlock(name="gps")
result = block.infer(GPSInput(
    op="solve",
    states=["start"],
    goal="goal",
    operators_json='[{"name":"step","preconditions":["start"],"add_list":["goal"],"delete_list":[]}]',
))
# result.ok → True; result.value → GPSOutput with solution (operator sequence)

Use capabilities, describe_operator_schema, and describe_store_schema before constructing new workflows. GPS is a deterministic symbolic solver: it can expose circular goals, missing goal-reducing operators, max-depth failures, optional-backend absence, and persistence fallback, but it does not infer whether the symbolic states capture the user's true intent or economic value.

Works well with: cog_arch_soar, solver, goal_engine

Public API

CogArchGPSBlock(AIBlock[GPSInput, GPSOutput, dict])

General Problem Solver cognitive architecture block.

Field Type Default
name str 'cog_arch_gps'
state dict \| None None
resource_bounds Any None

Methods:

infer(data: GPSInput) -> Result[GPSOutput]

GPSInput(BaseModel)

Field Type Default
op str required
states list[str] Field(default_factory=list)
goal str ''
operators_json str '[]'
query str ''
problem_id str ''

GPSOutput(BaseModel)

Field Type Default
op str required
ok bool required
solution list[str] Field(default_factory=list)
operators list[dict] Field(default_factory=list)
message str ''
error str ''
data_json str ''
degraded bool False
degradation_reason str ''
completion_state str 'qualified-draft'
warning_card str ''
evidence dict Field(default_factory=dict)
code str ''
llm_used bool False
llm_attempted bool False
agentic_fallback str ''
unavailable_backends list[str] Field(default_factory=list)
using_memory_fallback bool False
persistence_status str 'ok'

MCP Tools

Operation Source
operator_create cog_arch_gps_mcp
operator_get cog_arch_gps_mcp
operator_update cog_arch_gps_mcp
operator_list cog_arch_gps_mcp
operator_delete cog_arch_gps_mcp
problem_define cog_arch_gps_mcp
problem_get cog_arch_gps_mcp
problem_encode_llm cog_arch_gps_mcp
solve cog_arch_gps_mcp
solve_step cog_arch_gps_mcp
solve_status cog_arch_gps_mcp
solve_abort cog_arch_gps_mcp
solve_history cog_arch_gps_mcp
llm_suggest_operator cog_arch_gps_mcp
llm_explain_solution cog_arch_gps_mcp
episodic_store cog_arch_gps_mcp
episodic_recall cog_arch_gps_mcp
episodic_list cog_arch_gps_mcp
episodic_clear cog_arch_gps_mcp
verify_solution cog_arch_gps_mcp
find_analogy cog_arch_gps_mcp
realtime_bridge cog_arch_gps_mcp
hat_gate cog_arch_gps_mcp
gps_search cog_arch_gps_mcp
gps_info cog_arch_gps_mcp
list_patterns cog_arch_gps_mcp
capabilities cog_arch_gps_mcp
describe_store_schema cog_arch_gps_mcp
describe_operator_schema cog_arch_gps_mcp