pelesh · Oct 8, 2026 · Oct 7, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -7,6 +7,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [develop]
 
+## [2.1]
+
+### General
+- Fixed bug with strings in PETSc allocated strctures (#61)
+- Update dependencies to PETSc >=3.25 and ROCm >= 6.4 (#70)
+
+### User Interface
+- Add AgentiGrid experimental agentic AI user interface (#65)
+- Refactored visualizer (#69)
+
+### TCOPFLOW
+- Fix ramp parsing and GENRAMP indexing in TCOPFLOW (#62)
+- Improve preallocation of sparse matrices in TCOPFLOW (#71)
+
+
 ## [2.0]
 
 ### General

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -6,7 +6,7 @@ Preamble
 
 cmake_minimum_required(VERSION 3.18)
 
-project(ExaGO VERSION "2.0.0")
+project(ExaGO VERSION "2.1.0")
 
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/lib")

diff --git a/README.md b/README.md
@@ -32,11 +32,8 @@ Additionally, note that SCOPFLOW and SOPFLOW with HiOp solver use Ipopt to solve
 
 Detailed installation instructions are given at [INSTALL.md](./INSTALL.md) for information on acquiring, building and installing ExaGO.
 
-If you are a developer with access to the project, we also provide public binaries that are generated through our GitHub actions workflows documented in [README.md](.github/workflows/README.md), and with documentation about usage in the packages section of our repository. Check out a short (< 60s) demo of pulling down a version of ExaGO:
+If you are a developer with access to the project, we also provide public binaries that are generated through our GitHub actions workflows documented in [README.md](.github/workflows/README.md), and with documentation about usage in the packages section of our repository.
 
-[![asciicast](
-https://asciinema.org/a/KCi5TmUXc6zWDj7JYHzfSFxmw.png)](
-https://asciinema.org/a/KCi5TmUXc6zWDj7JYHzfSFxmw)
 
 ## Developer Guide
 
@@ -111,8 +108,8 @@ the guidelines are ambiguous or you have a particular question.
 ## Authors
 ExaGO<sup>TM</sup> was designed and implemented by Shrirang Abhyankar and has
 received significant contributions from Slaven Peles, Nicholson Koukpaizan,
-Maksudul Alam, Asher Mancinelli, Cameron Rutherford, Joshua Hambrick, Philip
-Fackler, Eve Tsybina, Bruce Palmer, Jaelyn Litzinger, William Perkins, Sayef
+Samim Konjicija, Maksudul Alam, Asher Mancinelli, Cameron Rutherford, Joshua Hambrick, Philip
+Fackler, Eve Tsybina, Shaked Regev, Bruce Palmer, Jaelyn Litzinger, William Perkins, Sayef
 Azad Sakin, Joseph Macam, and Ryan Danehy.
 
 ## Acknowledgement

diff --git a/agentigrid/agentigrid/engine/agent_loop.py b/agentigrid/agentigrid/engine/agent_loop.py
@@ -916,7 +916,7 @@ def _build_extra_args(self) -> list[str] | None:
         return args if args else None
 
     def _normalize_sopflow_wind_base(self) -> None:
-        """REVIEW (Slaven sign-off): model base-case wind as curtailable for SOPFLOW.
+        """REVIEW: model base-case wind as curtailable for SOPFLOW.
 
         case_ACTIVSg200 ships wind generators as must-run (Pmin = Pmax = nameplate),
         so any scenario whose wind availability is below nameplate is infeasible in
@@ -1000,7 +1000,7 @@ def run(self, base_case: Path, goal: str) -> SearchSession:
         logger.info("Parsing base case: %s", base_case)
         self._base_network = parse_matpower(base_case)
         self._current_network = self._base_network
-        # REVIEW (Slaven sign-off): model base-case wind as curtailable for SOPFLOW
+        # REVIEW: model base-case wind as curtailable for SOPFLOW
         # so scenarios with sub-nameplate wind are feasible (see the method + the
         # search.sopflow_curtailable_wind_base flag). No-op for non-SOPFLOW / disabled.
         self._normalize_sopflow_wind_base()

diff --git a/agentigrid/agentigrid_architecture.md b/agentigrid/agentigrid_architecture.md
@@ -7,8 +7,7 @@
 | Project | AgentiGrid (in process of integration into ExaGO) |
 | Version | 0.3 — PFLOW Direct Optimization |
 | Date | May 2026 |
-| Author | Samim K., University of Sarajevo |
-| Supervisor | Slaven P., Oak Ridge National Laboratory |
+| Author | Samim Konjicija, University of Sarajevo |
 | Target Platform | Linux (OpenSUSE Leap 15.6), Python 3.10+ |
 
 ---

diff --git a/agentigrid/claude_code_changes.md b/agentigrid/claude_code_changes.md
@@ -32,7 +32,7 @@ Confirmed `_augment_generator_mutation` already pins `Qmin == Qmax` (forced, not
 
 Prompt 13 re-ran an identical sweep 3× (50,504 tokens), prompt 17 2×.
 - `agentigrid/prompts/system_prompt.py`: explicit rule "DO NOT RE-RUN AN IDENTICAL SWEEP".
-- `agentigrid/engine/agent_loop.py` (deterministic backstop): session-level `_sweep_signature_cache` keyed by `_sweep_cache_key(data)` (mode + entity + mutation + candidate_set + feasibility + metric + predicate + dispatchable/cost params; excludes description/reasoning). An identical re-request is served via `_serve_cached_sweep` (journaled "cached — not re-solved", no re-solve); changed parameters → different signature → still executes. Soft guard, never blocks a changed sweep. **Decision flagged for Slaven:** nudge-only vs. nudge + cache (both implemented here).
+- `agentigrid/engine/agent_loop.py` (deterministic backstop): session-level `_sweep_signature_cache` keyed by `_sweep_cache_key(data)` (mode + entity + mutation + candidate_set + feasibility + metric + predicate + dispatchable/cost params; excludes description/reasoning). An identical re-request is served via `_serve_cached_sweep` (journaled "cached — not re-solved", no re-solve); changed parameters → different signature → still executes. Soft guard, never blocks a changed sweep. **Decision flagged for further discussion:** nudge-only vs. nudge + cache (both implemented here).
 
 ### Tests — `tests/test_sweep_corrections.py` (15)
 
@@ -123,7 +123,7 @@ A **named registry of verified primitives** (Phase 1; not free-form LLM code), e
 
 C2 generator primitive (dispatchable bounds + non-zero cost, fixed-injection regression, explicit coeffs, median mid-merit, fallback); registry (defaults, needs-base, direction, unknown-name raises, extensibility); `max_delta_v` (true max abs diff, max bus ≠ candidate bus, None without base); `reactive_adequacy` (adequate, inadequate-with-reason, non-convergence); standard predicate unchanged; `_augment_generator_mutation` (dispatchable resolution, explicit-strategy error, Q-pin, non-generator untouched); sweep-handler integration (unknown metric/predicate clean error, metric recorded, **default payload unchanged**, dispatched-Pg recorded). Full suite: **868 passed, 3 skipped**.
 
-### Decisions flagged for Slaven (non-blocking)
+### Decisions flagged for further discussion (non-blocking)
 
 1. Added-generator cost curve: case-median (default) vs an explicit typical mid-merit curve — drives the prompt-13 ranking entirely.
 2. Prompt-15 switched-in load size (`switched_load_mw`) — the voltage step scales with it.
@@ -173,7 +173,7 @@ For each candidate bus, a bisection on injection magnitude finds the largest inj
 
 System PF + PF-spec; candidate mutation on real ACTIVSg200 (PF ray, generator fixed injection, base untouched); binding identification (thermal/voltage with element, type-only fallback); bisection algorithm via synthetic oracle (boundary within tol, probe bound, cap-not-reached, first-probe-infeasible, convergence-as-infeasible); token-bounded boundary view (full vs summary, top-N bound, descending rank, 2000-bus size bound, grouped undetermined, threshold 0); `map_callables` (index order, exception capture, progress count); `_handle_boundary_sweep` orchestration (journals `max_feasible_mw`, one-LLM-turn invariant, base-infeasible abort, invalid-entity reject). Full suite: 836 passed, 3 skipped.
 
-### Decisions flagged for Slaven (non-blocking)
+### Decisions flagged for further discussion (non-blocking)
 
 1. Generator reactive handling: `Qg` free within `±0.4·ΔP` (current default) vs fixed at 0 — affects voltage-limited hosting numbers.
 2. PF source for the load prompt: system-average (default) vs each bus's own PF vs a stated PF.

diff --git a/agentigrid/launcher/ARCHITECTURE.md b/agentigrid/launcher/ARCHITECTURE.md
@@ -2,7 +2,7 @@
 
 **Version:** 1.0
 **Date:** 2026-03-27
-**Authors:** Samim / Claude (collaborative design)
+**Authors:** Samim Konjicija / Claude (collaborative design)
 
 ---
 

diff --git a/agentigrid/setup.py b/agentigrid/setup.py
@@ -13,7 +13,7 @@
     description="LLM-driven iterative simulation and analysis tool for ExaGO",
     long_description=long_description,
     long_description_content_type="text/markdown",
-    author="Slaven Peles",
+    author="Samim Konjicija",
     python_requires=">=3.10",
     packages=find_packages(exclude=["tests"]),
     install_requires=[

diff --git a/docs/AgentiGrid_modification_plan.md b/docs/AgentiGrid_modification_plan.md
@@ -163,7 +163,7 @@ Slice 1 (✅ done): add-entity primitives + binary-feasibility sweep + sweep rep
 
 ---
 
-## Open decisions for you / Slaven
+## Open decisions
 
 1. Confirm slice ordering — in particular whether B.1 (token view) jumps ahead of C1, given it gates the scaling half of the paper.
 2. Prompt 18 (min hot reserve under N-1) is the heaviest single item (SCOPFLOW + custom minimization) — is it in-scope for this paper, or deferred?

diff --git a/docs/manual/ExaGO_Manual.pdf b/docs/manual/ExaGO_Manual.pdf
diff --git a/docs/manual/main.tex b/docs/manual/main.tex
@@ -89,7 +89,7 @@
 %\renewcommand{\Re}{\R}
 %\newcommand{\Comment}[1]{}
 
-\newcommand{\exagoversion}{2.0.0}
+\newcommand{\exagoversion}{2.1.0}
 
 % Sets and accessors
 \newcommand{\Jgen}{J^\mathrm{gen}}

diff --git a/ipopt_ref.opt b/ipopt_ref.opt